Extra video on figuring out journal credibility: https://youtu.be/xCcNz14MPRc

f you need a refresher on the three criteria for establishing causality (association, temporal order, non-spuriousness), here's a foundational walkthrough: https://youtu.be/L40AEdRvWxU. And if you'd like to see how the criteria get applied to real experimental studies, here's a walk-through of three: https://youtu.be/GaoRbiROyEw

Now's a good moment to think about whether 251 is enough. The authors don't talk about how they decided on this sample size, but in well-designed experiments you'll often see a section called a power analysis that explains the reasoning behind sample size choices. This study doesn't include one — but you should know what it is because you'll see it in other papers and on the quiz. A power analysis is a calculation researchers do BEFORE collecting data to figure out how many participants they'll need. The calculation depends on three things:

Effect size — how big is the relationship you expect to find? Small effects need bigger samples to detect; large effects can be detected with smaller samples. Researchers usually estimate effect size from prior research on similar topics. Significance threshold (alpha) — typically set at .05. This is your tolerance for false positives — concluding there's an effect when there isn't. Statistical power (1 − beta) — typically set at .80. This is your tolerance for false negatives — failing to detect a real effect when one exists. A power of .80 means you have an 80% chance of finding a real effect if one is truly there.

Researchers plug these numbers into a tool like G*Power (a free software package) and it tells them the minimum sample size they need. Why does this matter? Because a study with too few participants can fail to detect real effects. When you see null results in a study (like H4, H6b in this article), one possible explanation is that the manipulation didn't work. Another is that the sample was too small to detect a small but real effect. Without a power analysis, we can't easily tell which is happening. In this study, dividing 251 participants across 6 conditions in the 3×2 factorial design gives roughly 42 participants per cell. A common rule of thumb for between-subjects experiments is at least 30 participants per cell — so this study clears that bar. But for detecting smaller effects, more would have been better. When you read a different experimental study on the quiz, look for whether the authors did a power analysis. If they did, it's a sign of methodological rigor. If they didn't, that's worth flagging as a limitation. Extra video on sample size in experiments: https://www.youtube.com/watch?v=v-dyn6tO5dQ

Make sure you're keeping two things straight that students mix up constantly: random assignment and random selection. Random selection (also called random sampling) is about who gets into the study in the first place. It happens at the recruitment stage. You start with a defined population (say, all U.S. tweens ages 8–13), and you use a random procedure to select some subset of them to invite. True random selection is rare in social science research because it requires a sampling frame — a complete list of every member of the population — which usually doesn't exist. Random selection is what supports external validity (generalizability of findings to the population).Random assignment is about what condition each participant ends up in once they're already in the study. It happens after recruitment. Every participant has an equal chance of being placed into any of the experimental conditions.

Random assignment is what supports internal validity — specifically, the non-spuriousness criterion for causality — because it equalizes pre-existing differences between groups so that when you see a difference on the DV, you can be confident it's because of the manipulation rather than because of some baseline group difference.A study can have one without the other. This study has random assignment (Qualtrics randomly placed each tween into one of 6 conditions) but not random selection (parents were recruited from Dynata's online panel, which is a convenience sample of people who voluntarily signed up to take surveys for points). That's why the authors can make causal claims internally but worry about generalizability in their limitations section. So when you answer this question, be specific: how were participants assigned, and was that the same as how participants were recruited? Those are different stages. Extra video on random assignment vs. random selection: https://www.youtube.com/watch?v=wB9S2od1wo0

H4: Sponsorship cue (manipulated IV: non-sponsored vs. sponsored) → perceived informative intent (DV)

The authors report no conflicts. What would be a hypothetical conflict of interest that would make you look more critically at the findings? For this study, examples might include: - Industry funding from a brand featured in the study. If Waterpik (or any other oral health product manufacturer) had funded this research, you'd want to look closely at whether the findings happen to favor their commercial interests — for example, did the authors find purchase intention was high across all conditions, suggesting unboxing videos are still effective marketing tools? - Funding from advertising or marketing trade groups. A grant from an industry group representing influencer marketers, advertising agencies, or social media platforms could create pressure to find that current practices are "fine" or that proposed regulations are unnecessary. - Funding from advocacy groups with strong prior positions. A grant from a children's media literacy advocacy organization could create pressure to find that training works well, since that aligns with their mission. Even well-intentioned advocacy funding can create subtle bias toward findings that support the funder's agenda. - Affiliations with platforms being studied. If an author had served as a paid consultant to YouTube, TikTok, or another platform that streams unboxing content, their interpretation of policy implications might be more sympathetic to platform-led solutions than to government regulation. - Authorship of competing measures or training programs. If an author had developed and licensed their own commercial advertising literacy curriculum, they'd have an incentive to find that THEIR approach works, or that brief in-app interventions (like the one tested here) are inadequate compared to fuller programs they sell. - Personal financial stakes in influencer content. An author who personally produces or monetizes children's online content would have a clear incentive in either direction depending on what they want the findings to show.

These would create potential bias — not necessarily making the research wrong, but something to be aware of. Note: a conflict of interest is about financial or professional incentives that could bias research, not about the researcher's personal identity or positionality. The fact that an author is or isn't a parent, for example, is not a conflict of interest. Notice that the funding section of this article is not present — the authors don't mention any external grant support. Sometimes the absence of external funding is itself informative. It may mean the research was done with internal university resources, which typically carries fewer strings than industry or advocacy funding. The fact that one of the authors is affiliated with the Joan Ganz Cooney Center at Sesame Workshop (a nonprofit research center focused on children's media) is worth noticing — that's not a conflict of interest, but it does signal that the author works in an environment with strong opinions about children's media. Sesame Workshop's mission shapes what research questions seem worth asking, which is different from financial bias but still worth being aware of.

This is the sort of question you'll have to answer on the quiz: In this study, the author(s) said that there are no conflicts of interest. Thinking about this study, can you think of any potential conflicts of interest that would cause you to look more critically at the findings? For example, think of particular organizations that the author(s) could be affiliated with or research funding sources that would lead you to think more about the credibility of the author and the findings. This is all hypothetical.

When they say "survey instrument" here they mean the questionnaire. This is not a survey study.

On the quiz you'll be asked to answer a question like this: Now that you've read the entire study, what TYPE of experiment is this? Please draw upon what you’ve learned about the different types and explain why this experiment exemplifies the type that it is.

Don't just guess at this. Walk through the 5-step decision process from the Identifying Experiment Types infographic: Step 1: How many things are you testing? If it's one IV, it's a single-factor design. If it's multiple IVs manipulated together, it's a factorial design. Count your IVs and multiply their levels to get the number of conditions. Step 2: How are participants placed into groups?

Random assignment → true experiment No random assignment, using existing groups (like classrooms) → quasi-experiment One group, no comparison → pre-experiment

Step 3: When do you measure?

Just once, after treatment → posttest only Before and after treatment → pretest-posttest Many times over time → time series

Step 4: Where does it happen?

Lab/controlled → lab experiment Real-world → field experiment Studying something that already happened in the world → natural experiment

Step 5: Who experiences what?

Same people experience all conditions → within-subjects Different people in each condition → between-subjects (most common)

Now apply each step to this study and put together the full label. The authors actually tell you part of the answer in the abstract — "3 (sponsored; non-sponsored; sponsorship unaddressed cue) x 2 (advertising training; no advertising training) randomized experimental design" — but that doesn't capture everything. Add the missing pieces from steps 3, 4, and 5. A complete answer should sound something like: "this is a [factorial / single-factor] [true / quasi / pre] [posttest-only / pretest-posttest] [lab / field / natural / online] [between-subjects / within-subjects] experiment, because [walk through your reasoning]." The Experiment Types Overview infographic in Module 7 has the full taxonomy with definitions for each design type.

Like perceived selling intent, this is a measured variable, and it does double duty across the study — IV in some hypotheses, DV in others. Operationalization: Three survey items adapted from Shan et al. (2020), each on a 5-point response scale (1: definitely not to 5: definitely yes). The items asked whether the boy in the video (1) wants to help others by sharing useful information about healthy teeth, (2) wants to help others by sharing useful information about the water flosser, and (3) thinks the water flosser is good to use. The three items were averaged into a scale. Level of measurement: Interval. The 5-point Likert-type scale is conventionally treated as interval in social science research. Use: Has multiple roles across the study:

Dependent variable in H4 (non-sponsored video → perceived informative intent), H5 (unaddressed video → perceived informative intent), and H6b (training → perceived informative intent in non-sponsored condition) Independent variable in H2 (perceived informative intent → educational recall) and RQ2a/RQ2b (perceived intent → recall, purchase intention)

Was it manipulated? No — this is a key point. Perceived informative intent was measured via survey items, not manipulated. The researchers didn't make some kids perceive informative intent and others not; they just asked all kids how much informative intent they perceived. This means that whenever perceived informative intent is treated as an "IV" (in H2 and the RQs), the causality claims are weaker than they would be for a manipulated variable. Measurement validity/reliability: Cronbach's alpha was .82, which is good (above the .70 threshold for acceptable internal consistency, and well above the .80 threshold for good consistency). The three items reliably measure a single underlying construct. Compare this to perceived selling intent, which has only 2 items and reports a correlation (r = .53) instead of alpha. The 3-item informative intent scale is more reliable than the 2-item selling intent scale — both because more items generally produce more stable measurement, and because alpha .82 is stronger than r = .53. Conceptual definition is on page 275, embedded in the schema theory discussion: "activation of an informative/educational schema likely directs the viewer's attention to educational information unrelated to the product's features." So informative intent is conceptualized as the recognition that content is meant to be objective or educational rather than commercial.

Two recall measures are bundled under this variable name — product recall and health recall — but they're scored separately and used as separate DVs in the analysis. Operationalization: Four multiple-choice items total, two for each recall type:

Product recall (2 items): How much water could the device hold? How many water-pressure speed options does it have? Health recall (2 items): What is enamel? What is plaque?

Each item was scored as correct (1) or incorrect (0). The two product items were summed for a 0–2 product recall score; the two health items were summed for a 0–2 health recall score. Level of measurement: Each item is nominal (correct or incorrect). The summed score (0, 1, or 2) is technically ordinal — there's order (more correct is better) but the intervals between scores aren't necessarily meaningful in a strict measurement sense. Some researchers would treat this as interval for analysis purposes. Use: Dependent variable in H1 (perceived selling intent → product recall), H2 (perceived informative intent → health recall), and RQ2a (perceived intent × training → recall). Measurement validity: Worth flagging that these are multiple-choice recognition measures, not free recall. Recognition is generally easier than recall — kids might pick the right answer from four options without genuinely having retained the information. A more rigorous measure would have asked open-ended questions ("How much water could the device hold?") and scored kids' written or spoken answers. The recognition format probably inflates scores compared to free recall, which connects to the ceiling effect the authors flag in their limitations. Also worth noting: with only TWO items per recall type, you can't compute meaningful internal consistency. The measures are short for practical reasons (kids' attention spans, survey length), but it's a measurement weakness worth acknowledging. Conceptual definition: The article doesn't provide an explicit conceptual definition of "content recall." The authors describe what they're measuring (memory of product attributes vs. memory of health information) but don't articulate the underlying theoretical construct. Worth noting as a #conceptualdefinition gap.

This is arguably the most consequential DV in the study — purchase intention connects directly to real-world consumer behavior — but it's measured with the lightest possible touch. Operationalization: A single survey item: "Will you ask your parents to buy a [company] water flosser?" Response options ranged from 1 (definitely not) to 5 (definitely yes). Level of measurement: Interval (5-point Likert-type scale, treated conventionally as interval). Use: Dependent variable in RQ2b (perceived intent × training → purchase intention). It also appears in the path analysis as an outcome of perceived intent across multiple analyses. Measurement validity/reliability: This is where it gets concerning. Single-item measurement means:

Reliability is unmeasurable. You need at least two items to compute internal consistency. With one item, we have no statistical evidence the measure is reliable. Validity rests entirely on this question. If the wording isn't quite right — and "ask your parents to buy" might capture something different from underlying purchase desire — there's no way to triangulate against a second item. Variability is constrained. Five options for 251 kids will produce clusters at certain values, limiting the analysis's power to detect subtle effects.

Researchers sometimes use single-item measures for face-valid, simple constructs, especially when surveying children where survey length is a real constraint. But for a key DV in an experimental study, a multi-item scale would have been stronger. This is a measurement weakness the authors don't acknowledge in their limitations. Conceptual definition: The article doesn't provide an explicit conceptual definition of "purchase intention." The construct is implied — kids' interest in acquiring the depicted product — but never theoretically defined. Like content recall, this is a #conceptualdefinition gap worth flagging.

This is a measured variable — and that's important to understand because perceived intent shows up in nearly every hypothesis but is never randomly assigned. Operationalization: Two survey items adapted from Rozendaal et al. (2016), each on a 5-point response scale (1: definitely not to 5: definitely yes). The items asked whether the boy in the video (1) is being paid by the company, and (2) is trying to get me to buy a water flosser. The items were averaged into a scale. Level of measurement: Interval. The 5-point Likert-type scale is conventionally treated as interval in social science research, even though strictly speaking it's debatable. Use: Has multiple roles across the study:

Dependent variable in H3 (sponsored video → perceived selling intent) and H6a (training → perceived selling intent in sponsored condition) Independent variable in H1 (perceived selling intent → product recall) and RQ2a/RQ2b (perceived intent → recall, purchase intention)

Was it manipulated? No — this is a critical point. Perceived selling intent was MEASURED via survey items, not manipulated. The researchers didn't make some kids perceive selling intent; they just asked all kids how much selling intent they perceived. This means that whenever perceived selling intent is treated as an "IV" (in H1 and the RQs), the causal claims are weaker than they would be for a manipulated variable. Measurement validity/reliability: The two items were significantly correlated (r = 0.53, p < .001), indicating moderate consistency between them. Note: with only two items, the authors report a correlation rather than Cronbach's alpha. Alpha is conventionally calculated for three or more items. A correlation of .53 isn't strong — it suggests the two items are tapping a common construct but capturing somewhat different aspects of it. Conceptual definition is on page 274, embedded in the description of the Persuasion Knowledge Model: "individuals will engage in cognitive coping strategies when they identify selling intent within a given message." So selling intent is conceptualized as the recognition that a message is trying to persuade.

In this hypothesis/research question, you'd be asked to identify the independent variable and the dependent variable. And there is a moderating variable in this hypothesis/research question. A moderating variable is a variable that affects the direction and/or strength of the relationship between an independent variable and a dependent variable. [Example: In a study examining the relationship between study time and exam performance, motivation level acts as a moderating variable. For highly motivated students, more study time leads to significantly better exam performance, while for lowly motivated students, the effect of study time on performance is much weaker. Thus, motivation changes the strength of the relationship between study time and exam performance, demonstrating its role as a moderating variable.] So for this hypothesis or research question, you will need to list the independent variable, the moderating variable, and the dependent variable. Read closely to see what the author(s) argue is happening here.

If you need more help with moderators/mediators, here’s an extra video: https://www.youtube.com/watch?v=FzM0_GC082A

In this hypothesis/research question, you'd be asked to identify the independent variable and the dependent variable. And there is a moderating variable in this hypothesis/research question. A moderating variable is a variable that affects the direction and/or strength of the relationship between an independent variable and a dependent variable. [Example: In a study examining the relationship between study time and exam performance, motivation level acts as a moderating variable. For highly motivated students, more study time leads to significantly better exam performance, while for lowly motivated students, the effect of study time on performance is much weaker. Thus, motivation changes the strength of the relationship between study time and exam performance, demonstrating its role as a moderating variable.] So for this hypothesis or research question, you will need to list the independent variable, the moderating variable, and the dependent variable. Read closely to see what the author(s) argue is happening here.

If you need more help with moderators/mediators, here’s an extra video: https://www.youtube.com/watch?v=FzM0_GC082A

This is more of a descriptive RQ, so no IV and DV per se, but sort of could be considered like RQ1: Perceived informative intent + perceived selling intent (both measured IVs, compared) → purchase intention (DV).

This isn't a university or a college, so it should cause you to think more about the author's affiliation. Research what this affiliation is and if/how it impacts your credibility assessment of the author.

In this hypothesis/research question, you'd be asked to identify the independent variable and the dependent variable. And there is a moderating variable in this hypothesis/research question. A moderating variable is a variable that affects the direction and/or strength of the relationship between an independent variable and a dependent variable. [Example: In a study examining the relationship between study time and exam performance, motivation level acts as a moderating variable. For highly motivated students, more study time leads to significantly better exam performance, while for lowly motivated students, the effect of study time on performance is much weaker. Thus, motivation changes the strength of the relationship between study time and exam performance, demonstrating its role as a moderating variable.] So for this hypothesis or research question, you will need to list the independent variable, the moderating variable, and the dependent variable. Read closely to see what the author(s) argue is happening here.

If you need more help with moderators/mediators, here’s an extra video: https://www.youtube.com/watch?v=FzM0_GC082A

Now that you’ve read the entire study, you should always make sure that you can state what were the primary findings of the research. You don’t have to give a bunch of details. This should be 1-4 sentences expressing the big takeaway that the researchers found answering their hypotheses/research questions.

H6a: Training video (manipulated IV: training vs. no training) → perceived selling intent (DV) — within the sponsored condition specifically. They're arguing for an interaction effect here.

When you're working on these quizzes, you'll be asked to find limitations that the author(s) discusses. Remember that limitations are not the research findings, but rather things that the author(s) wishes that they had done differently or challenges with the sample, measurement, procedures, etc. that could impact the results.

Working through causality questions can feel overwhelming, so here's a framework. For each hypothesis, walk through the three criteria one at a time:

Association — Did the authors find a statistical relationship between the IV and DV in this specific hypothesis? Look at the results section. Was the hypothesis supported, partially supported, or unsupported? Temporal order — Did the IV come before the DV in time? In a true experiment with random assignment, the manipulation always comes first by design. But watch for cases where the "IV" was actually MEASURED rather than manipulated — those weaken the temporal order claim because measurement and outcome happen close together in the survey. Non-spuriousness — Did the design rule out third variables? Random assignment is the main tool here. If participants were randomly assigned to conditions, pre-existing differences should be roughly equal across groups. If the IV was measured (not manipulated), there's no random assignment to that variable, and confounds become a real concern.

Watch out: this study has a mix of MANIPULATED variables (sponsorship cue, training video — both randomly assigned) and MEASURED variables (perceived selling intent, perceived informative intent). Some hypotheses test relationships involving measured variables, where causality claims are weaker even though the overall study uses random assignment. For a walk-through of how to apply these criteria to real experimental studies (3 examples): https://youtu.be/GaoRbiROyEw For a longer, more detailed in-class version: https://www.youtube.com/watch?v=xcItkXe9G6E

This is the second manipulated independent variable, and it's also doing double duty as a moderator in the path analysis. Operationalization: A 90-second animated video featuring two cartoon spokespeople (a 12-year-old male and an undergraduate female) explaining that some YouTube videos are designed to sell products. The video describes "influencer" tactics, warns viewers they can be "tricked," and lists cues to watch for ("sponsor," "partnership"). Half of the participants saw this video before the unboxing video; the other half went straight to the unboxing video. Level of measurement: Nominal. Two categories: training vs. no training. Use: Independent variable in H6a (training → detection of selling intent in sponsored content) and H6b (training → detection of informative intent in non-sponsored content). Also serves as a moderator in RQ2a and RQ2b — the authors test whether training changes the relationship between perceived video intent and recall (RQ2a) or purchase intention (RQ2b). How was it manipulated? Qualtrics randomly assigned half of participants to view the training video before the experimental unboxing video. The other half saw the unboxing video alone. Random assignment is what allows the authors to make causal claims about the training's effects. Measurement validity: The training video plausibly teaches what advertising literacy is, but with no manipulation check, we can't verify whether kids absorbed the content. The authors' own discussion notes this ambiguity: they aren't sure whether the training "truly taught the advertising literacy content, or [cued] the use of advertising literacy skills they already had." Conceptual definition is on page 278: "'advertising literacy.' Advertising literacy skills help children filter the information they see in advertisements and view it through a critical lens to determine the intent behind it." The training video is meant to build or activate exactly these skills.

Working through causality questions can feel overwhelming, so here's a framework. For each hypothesis, walk through the three criteria one at a time:

Association — Did the authors find a statistical relationship between the IV and DV in this specific hypothesis? Look at the results section. Was the hypothesis supported, partially supported, or unsupported? Temporal order — Did the IV come before the DV in time? In a true experiment with random assignment, the manipulation always comes first by design. But watch for cases where the "IV" was actually MEASURED rather than manipulated — those weaken the temporal order claim because measurement and outcome happen close together in the survey. Non-spuriousness — Did the design rule out third variables? Random assignment is the main tool here. If participants were randomly assigned to conditions, pre-existing differences should be roughly equal across groups. If the IV was measured (not manipulated), there's no random assignment to that variable, and confounds become a real concern.

Watch out: this study has a mix of MANIPULATED variables (sponsorship cue, training video — both randomly assigned) and MEASURED variables (perceived selling intent, perceived informative intent). Some hypotheses test relationships involving measured variables, where causality claims are weaker even though the overall study uses random assignment. For a walk-through of how to apply these criteria to real experimental studies (3 examples): https://youtu.be/GaoRbiROyEw For a longer, more detailed in-class version: https://www.youtube.com/watch?v=xcItkXe9G6E

H6b: Training video (manipulated IV: training vs. no training) → perceived informative intent (DV) — within the non-sponsored condition specifically.

In this hypothesis/research question, you'd be asked to identify the independent variable and the dependent variable. And there is a moderating variable in this hypothesis/research question. A moderating variable is a variable that affects the direction and/or strength of the relationship between an independent variable and a dependent variable. [Example: In a study examining the relationship between study time and exam performance, motivation level acts as a moderating variable. For highly motivated students, more study time leads to significantly better exam performance, while for lowly motivated students, the effect of study time on performance is much weaker. Thus, motivation changes the strength of the relationship between study time and exam performance, demonstrating its role as a moderating variable.] So for this hypothesis or research question, you will need to list the independent variable, the moderating variable, and the dependent variable. Read closely to see what the author(s) argue is happening here.

If you need more help with moderators/mediators, here’s an extra video: https://www.youtube.com/watch?v=FzM0_GC082A

Working through causality questions can feel overwhelming, so here's a framework. For each hypothesis, walk through the three criteria one at a time:

Association — Did the authors find a statistical relationship between the IV and DV in this specific hypothesis? Look at the results section. Was the hypothesis supported, partially supported, or unsupported? Temporal order — Did the IV come before the DV in time? In a true experiment with random assignment, the manipulation always comes first by design. But watch for cases where the "IV" was actually MEASURED rather than manipulated — those weaken the temporal order claim because measurement and outcome happen close together in the survey. Non-spuriousness — Did the design rule out third variables? Random assignment is the main tool here. If participants were randomly assigned to conditions, pre-existing differences should be roughly equal across groups. If the IV was measured (not manipulated), there's no random assignment to that variable, and confounds become a real concern.

Watch out: this study has a mix of MANIPULATED variables (sponsorship cue, training video — both randomly assigned) and MEASURED variables (perceived selling intent, perceived informative intent). Some hypotheses test relationships involving measured variables, where causality claims are weaker even though the overall study uses random assignment. For a walk-through of how to apply these criteria to real experimental studies (3 examples): https://youtu.be/GaoRbiROyEw For a longer, more detailed in-class version: https://www.youtube.com/watch?v=xcItkXe9G6E

Working through causality questions can feel overwhelming, so here's a framework. For each hypothesis, walk through the three criteria one at a time:

Association — Did the authors find a statistical relationship between the IV and DV in this specific hypothesis? Look at the results section. Was the hypothesis supported, partially supported, or unsupported? Temporal order — Did the IV come before the DV in time? In a true experiment with random assignment, the manipulation always comes first by design. But watch for cases where the "IV" was actually MEASURED rather than manipulated — those weaken the temporal order claim because measurement and outcome happen close together in the survey. Non-spuriousness — Did the design rule out third variables? Random assignment is the main tool here. If participants were randomly assigned to conditions, pre-existing differences should be roughly equal across groups. If the IV was measured (not manipulated), there's no random assignment to that variable, and confounds become a real concern.

Watch out: this study has a mix of MANIPULATED variables (sponsorship cue, training video — both randomly assigned) and MEASURED variables (perceived selling intent, perceived informative intent). Some hypotheses test relationships involving measured variables, where causality claims are weaker even though the overall study uses random assignment. For a walk-through of how to apply these criteria to real experimental studies (3 examples): https://youtu.be/GaoRbiROyEw For a longer, more detailed in-class version: https://www.youtube.com/watch?v=xcItkXe9G6E

Working through causality questions can feel overwhelming, so here's a framework. For each hypothesis, walk through the three criteria one at a time:

Association — Did the authors find a statistical relationship between the IV and DV in this specific hypothesis? Look at the results section. Was the hypothesis supported, partially supported, or unsupported? Temporal order — Did the IV come before the DV in time? In a true experiment with random assignment, the manipulation always comes first by design. But watch for cases where the "IV" was actually MEASURED rather than manipulated — those weaken the temporal order claim because measurement and outcome happen close together in the survey. Non-spuriousness — Did the design rule out third variables? Random assignment is the main tool here. If participants were randomly assigned to conditions, pre-existing differences should be roughly equal across groups. If the IV was measured (not manipulated), there's no random assignment to that variable, and confounds become a real concern.

Watch out: this study has a mix of MANIPULATED variables (sponsorship cue, training video — both randomly assigned) and MEASURED variables (perceived selling intent, perceived informative intent). Some hypotheses test relationships involving measured variables, where causality claims are weaker even though the overall study uses random assignment. For a walk-through of how to apply these criteria to real experimental studies (3 examples): https://youtu.be/GaoRbiROyEw For a longer, more detailed in-class version: https://www.youtube.com/watch?v=xcItkXe9G6E

Please note that there is more information about the sample down here!

Working through causality questions can feel overwhelming, so here's a framework. For each hypothesis, walk through the three criteria one at a time:

Association — Did the authors find a statistical relationship between the IV and DV in this specific hypothesis? Look at the results section. Was the hypothesis supported, partially supported, or unsupported? Temporal order — Did the IV come before the DV in time? In a true experiment with random assignment, the manipulation always comes first by design. But watch for cases where the "IV" was actually MEASURED rather than manipulated — those weaken the temporal order claim because measurement and outcome happen close together in the survey. Non-spuriousness — Did the design rule out third variables? Random assignment is the main tool here. If participants were randomly assigned to conditions, pre-existing differences should be roughly equal across groups. If the IV was measured (not manipulated), there's no random assignment to that variable, and confounds become a real concern.

Watch out: this study has a mix of MANIPULATED variables (sponsorship cue, training video — both randomly assigned) and MEASURED variables (perceived selling intent, perceived informative intent). Some hypotheses test relationships involving measured variables, where causality claims are weaker even though the overall study uses random assignment. For a walk-through of how to apply these criteria to real experimental studies (3 examples): https://youtu.be/GaoRbiROyEw For a longer, more detailed in-class version: https://www.youtube.com/watch?v=xcItkXe9G6E

Working through causality questions can feel overwhelming, so here's a framework. For each hypothesis, walk through the three criteria one at a time:

Association — Did the authors find a statistical relationship between the IV and DV in this specific hypothesis? Look at the results section. Was the hypothesis supported, partially supported, or unsupported? Temporal order — Did the IV come before the DV in time? In a true experiment with random assignment, the manipulation always comes first by design. But watch for cases where the "IV" was actually MEASURED rather than manipulated — those weaken the temporal order claim because measurement and outcome happen close together in the survey. Non-spuriousness — Did the design rule out third variables? Random assignment is the main tool here. If participants were randomly assigned to conditions, pre-existing differences should be roughly equal across groups. If the IV was measured (not manipulated), there's no random assignment to that variable, and confounds become a real concern.

Watch out: this study has a mix of MANIPULATED variables (sponsorship cue, training video — both randomly assigned) and MEASURED variables (perceived selling intent, perceived informative intent). Some hypotheses test relationships involving measured variables, where causality claims are weaker even though the overall study uses random assignment. For a walk-through of how to apply these criteria to real experimental studies (3 examples): https://youtu.be/GaoRbiROyEw For a longer, more detailed in-class version: https://www.youtube.com/watch?v=xcItkXe9G6E

Some journals — especially newer ones and those with applied audiences — include a section like this as a public scholarship outreach effort. It's a more accessible summary of the study, written in plain language, structured clearly (what we knew, what's new, why it matters), and stripped of jargon. Think of it as a built-in, author-approved version of what a science journalist would write. Notice the three-part structure here:

Prior State of Knowledge: what the field knew before this study Novel Contributions: what THIS study adds Practical Implications: why anyone outside academia should care

This is a really useful section to read first, before diving into the abstract. It tells you the gist of the study in language a non-specialist can follow. Now imagine a journalist covering this study. They might write a headline like: "YouTube unboxing videos confuse kids — but a 90-second warning helps" "Tweens see sponsored YouTube videos as both ads AND honest reviews, study finds" "Even with disclosures, kids struggle to spot YouTube ads — researchers say platforms should help" Notice what those headlines do: they translate the academic findings into something concrete and actionable. They emphasize the practical takeaway (kids need help, training works briefly) rather than the technical findings (path analysis showed moderation by training condition, p < .05). That translation is part of what makes research findings reach audiences beyond other researchers. Not every journal does this. Older and more traditional journals tend to skip the impact summary entirely. When you DO see one, take advantage — it's a free roadmap to the study.

Mediation analysis is used to understand how or why one thing influences another by introducing a middle factor that links them together. Think of it as a “go-between.”

Working through causality questions can feel overwhelming, so here's a framework. For each hypothesis, walk through the three criteria one at a time:

Association — Did the authors find a statistical relationship between the IV and DV in this specific hypothesis? Look at the results section. Was the hypothesis supported, partially supported, or unsupported? Temporal order — Did the IV come before the DV in time? In a true experiment with random assignment, the manipulation always comes first by design. But watch for cases where the "IV" was actually MEASURED rather than manipulated — those weaken the temporal order claim because measurement and outcome happen close together in the survey. Non-spuriousness — Did the design rule out third variables? Random assignment is the main tool here. If participants were randomly assigned to conditions, pre-existing differences should be roughly equal across groups. If the IV was measured (not manipulated), there's no random assignment to that variable, and confounds become a real concern.

Watch out: this study has a mix of MANIPULATED variables (sponsorship cue, training video — both randomly assigned) and MEASURED variables (perceived selling intent, perceived informative intent). Some hypotheses test relationships involving measured variables, where causality claims are weaker even though the overall study uses random assignment. For a walk-through of how to apply these criteria to real experimental studies (3 examples): https://youtu.be/GaoRbiROyEw For a longer, more detailed in-class version: https://www.youtube.com/watch?v=xcItkXe9G6E

This is one of the two manipulated independent variables in the study. Let's walk through the analysis. Operationalization: Three different versions of an unboxing video, each about three minutes long, all featuring the same 12-year-old male actor demonstrating the same water flosser. The body of each video is identical. What differs is the framing at the beginning and end:

Non-sponsored: Boy explains it's a school project for "Mrs. Jenkins' 6th grade class." Sponsored: Boy explicitly thanks his "friends at [company]" who sent him the product and gives the price and retailers. Sponsorship unaddressed: Boy says he saw the product at a friend's house, mentions the price and availability, but never explicitly states a sponsor relationship.

Level of measurement: Nominal. The three conditions are categories without inherent order. (You could argue ordinal — non-sponsored → unaddressed → sponsored represents increasing levels of commercial intent — but the authors treat these as nominal categories in their analysis, so nominal is the safer answer.) Use: Independent variable in H3 (sponsored vs. non-sponsored → perceived selling intent), H4 (non-sponsored vs. sponsored → perceived informative intent), and H5 (unaddressed vs. non-sponsored → perceived informative intent). How was it manipulated? Qualtrics randomly assigned each child to one of the three video conditions. The video was streamed within Qualtrics from a private YouTube channel "in order to boost children's beliefs that these were real videos on the YouTube platform." So the manipulation involves both (a) which version of the video the child saw, and (b) the framing that made it feel like real YouTube content rather than a study stimulus. Measurement validity: The three conditions are differentiated by clear scripted statements (sponsorship explicit, school report explicit, or ambiguous). Face validity is decent — the videos plausibly represent the three real-world scenarios the authors care about. The bigger concern is that the same actor and same body content appears in all three, which controls for confounds but reduces ecological validity (real influencer videos vary far more in style, polish, and host). Conceptual definition is on page 273: "'influencers,' paid or otherwise compensated by marketers in exchange for favorable reviews." That's the underlying construct the sponsored condition is operationalizing.

Structural Equation Modeling (SEM) is a statistical technique in social sciences that allows us to examine complex relationships between multiple variables in a single, integrated model. It combines measurement models to estimate hidden or “latent” variables, like intelligence or happiness, using indicators we can observe, and structural models that show how these variables influence each other. SEM enables us to see not only direct effects but also indirect and mutual relationships, all at once. By doing this, SEM helps us test if our theoretical model accurately represents the data, offering a more complete view of the relationships between observed and unobserved factors.

We discussed online panels in the sampling module. You may want to refresh your understanding of them. https://youtu.be/Ywikc4_GL8w

Let's start with journal credibility. Journal of Children and Media (which we've seen before) is published by Routledge/Taylor & Francis, which is a well-established academic publisher. But sometimes established publishers aren't ALWAYS publishing credible journals. To verify credibility, you'd want to check tools we discussed in Module 5: SCImago Journal Rank (SJR), Ulrich's Web (to confirm peer-review status), or the journal's own website. This journal is indexed in major databases and has an ISSN, which are good signs. On the quiz, you won't have internet access, but you should know WHICH tools you'd use and WHAT you'd look for.

Path analysis is a statistical tool in social sciences that allows us to map and measure both direct and indirect relationships between multiple variables to understand how they influence a particular outcome. In this approach, each variable is represented as a node, and arrows (or "paths") between nodes show how one variable affects another. Some variables have direct effects, with an arrow pointing straight to the outcome, while others work indirectly, influencing an intermediate variable first. By drawing and analyzing these paths, path analysis helps us test theories about how variables interact to produce an outcome, giving us a detailed view of the connections and the strength of each relationship in the model.

Moderation analysis helps you find out when or for whom a relationship between two things is stronger or weaker by introducing a third factor, called a moderator. This moderator changes how the two main things relate to each other.

r is the correlation coefficient. We will cover this more in Module 9. This number that tells us how strongly two variables are related and in which direction. It ranges from -1 to +1. A positive value (closer to +1) means that as one variable increases, the other tends to increase as well - this is called a positive correlation. A negative value (closer to -1) means that as one variable increases, the other tends to decrease - this is a negative correlation. When r is close to 0, it means there is little to no relationship between the two variables.

Mediation analysis is used to understand how or why one thing influences another by introducing a middle factor that links them together. Think of it as a “go-between.”

Path analysis is a statistical tool in social sciences that allows us to map and measure both direct and indirect relationships between multiple variables to understand how they influence a particular outcome. In this approach, each variable is represented as a node, and arrows (or "paths") between nodes show how one variable affects another. Some variables have direct effects, with an arrow pointing straight to the outcome, while others work indirectly, influencing an intermediate variable first. By drawing and analyzing these paths, path analysis helps us test theories about how variables interact to produce an outcome, giving us a detailed view of the connections and the strength of each relationship in the model.

H3: Sponsorship cue (manipulated IV: sponsored vs. non-sponsored) → perceived selling intent (DV)

A 'between-subjects' design in social science experiments refers to a research setup where different groups of participants are exposed to different conditions or treatments. Each participant experiences only one condition, so comparisons are made between groups. For example, imagine you're testing the effectiveness of two teaching methods on student performance. In a between-subjects design, you would have one group of students using Method A and a different group using Method B. You'd then compare the results between these two groups to see which method works better. This type of design is useful because it helps to avoid carryover effects, where the experience of one condition might influence performance in another if the same participants experienced both. However, it also requires more participants because each person only provides data for one condition.

This study is a between-subjects design — different participants are in each condition. With 251 participants and 6 conditions, that means roughly 42 different kids per condition. Each kid sees ONE combination of sponsorship cue and training video. They don't see all 6. The alternative is a within-subjects design, where the same participants experience all (or several) conditions. If this study had been within-subjects, each kid would have watched all six versions of the video — sponsored with training, sponsored without training, non-sponsored with training, and so on — and the researchers would compare each kid's responses across the conditions they personally experienced. Why choose one over the other? Between-subjects advantages:

Cleaner — each participant only sees one condition, so there's no carryover from one to another. Avoids order effects — a participant doesn't get tired or bored partway through. Avoids demand characteristics that come from comparing conditions — when a kid sees both sponsored and non-sponsored videos back-to-back, they might guess what the study is about.

Between-subjects disadvantages:

Requires more participants — you need a separate group for each condition. Pre-existing differences between groups can muddy the results, which is why you need random assignment.

Within-subjects advantages:

Statistically more efficient — each participant serves as their own comparison, which controls for individual differences automatically. Requires fewer total participants.

Within-subjects disadvantages:

Carryover effects — what you saw first might affect your reaction to what comes next. Order effects — fatigue, practice, or boredom can shape later responses. Demand characteristics — participants may guess the study's purpose by comparing conditions.

For this study, between-subjects makes sense because: (1) the manipulations work better when kids encounter only one version (seeing all three sponsorship cues back-to-back would tip them off to the research question); (2) the kids are young and might struggle with multiple video viewings; (3) the authors had access to enough participants through Dynata to fill all 6 cells. When you read any experiment on the quiz, ask: did each participant experience all conditions (within-subjects), or just one (between-subjects)? You can tell by looking at how the conditions are described and how participants were assigned. Most experiments are between-subjects — it's the more common choice.

This is a video on understanding the hypotheses/research questions in THIS study: https://youtu.be/oayIhc18HEg

When you go through this question, work through the seven threats systematically rather than guessing. Here's the list, with how each applies — or doesn't — to a study like this one:

History — events outside the study that happen between measurements and influence the DV. Less of a concern in a short, single-session online experiment like this one, since there's no extended time window. But it could matter if, say, a major news story about influencer scams broke during the data collection period (May 27 – June 3, 2022).

Maturation — natural changes in participants over time. Mostly a concern for studies that span weeks or months. Tweens didn't age meaningfully during this 15-minute experiment.

Testing — taking a pretest changes posttest responses. Not relevant here because there's no pretest.

Instrumentation — the measurement tool changes between measurements. Not an issue with a posttest-only design.

Regression to the mean — extreme scorers drift toward the average on retesting. Not relevant without repeated measurement.

Selection bias — pre-existing differences between groups. This is where random assignment does its work. The authors used random assignment, which addresses selection — but only if the randomization actually produced equivalent groups. (Did the authors check for baseline equivalence on demographics? Look for that.)

Attrition — participants dropping out, especially if they drop out non-randomly. The authors note 13 children quit early and 3 declined to participate. Were dropouts equally distributed across conditions? If kids in the sponsored condition were more likely to quit than kids in the non-sponsored condition, that's a problem.

Then there's the broader category of demand characteristics / Hawthorne effect / experimenter effects, which threaten internal validity through participant behavior rather than through the seven classic threats. Worth asking: did the kids behave naturally, or were they performing for the researcher / their parent (who was likely nearby during this online study)?

Random assignment is a powerful tool — it addresses several of these threats at once — but it's not a magic eraser. Some threats survive randomization. Your job is to figure out which ones the authors handled well, and which ones they didn't address.

Extra video on mitigating threats to internal validity: https://youtu.be/3GW13A4-eSQ. The Threats to Internal Validity infographic in Module 7 has the full list with examples of each.

In-situ means in the original/normal place.

In these quizzes, you're going to always be asked about the bigger research question and purpose/goal of the study – not tied the hypotheses or research questions with variables – but the larger research question and objective. Ways that you can think about this: why bother doing this study? How does conducting this study add to our understanding of the world? How do the researchers justify spending time and money studying this question? [You don’t have to answer these questions directly, but these are the sort of things that you should think about with regard to the bigger research question and goal.] This will be in the introduction section of the study. [Please note that sometimes there is a heading "Introduction" and sometimes there is not - but the text after the abstract is the introduction.]

Here's a video about the design of THIS study: https://youtu.be/F2mVWHvj3KM

This is a bit more of a descriptive hypothesis, so no IV/DV. But roughly, H5: Sponsorship cue (manipulated IV: unaddressed compared to non-sponsored) → perceived informative intent (DV). Descriptive hypothesis comparing the unaddressed condition to the non-sponsored condition.

Remember that the discussion section summarizes what the author(s) found, how it is contextualized within the existing body of research, and what contribution this study's findings make to this area of research. Then authors usually speculate about some bigger picture questions related to the research. Authors then discuss limitations and opportunities for future research.

Extra video on eliminating alternative explanations: https://www.youtube.com/watch?v=euPS_SMYoU8

Video about COI in THIS study: https://youtu.be/pVWhgozTXXw

When you encounter a methods section, one thing you'll need to do — both for this study and on the in-class quiz — is figure out what TYPE of experiment it is. Don't just guess. Walk through the 5-step decision process from the Identifying Experiment Types infographic in Module 7. Step 1: How many things are you testing? Count the IVs and their levels.

One IV → single-factor design Multiple IVs → factorial design (multiply the levels: a 2×2 has 4 conditions, a 3×2 has 6, a 2×3 has 6, a 3×3 has 9, etc.)

For this study: two IVs (sponsorship cue with 3 levels, training video with 2 levels). 3 × 2 = 6 conditions. Factorial. Step 2: How are participants placed into groups?

Random assignment → true experiment No random assignment, using existing groups (like classrooms) → quasi-experiment One group, no comparison → pre-experiment

For this study: Qualtrics randomly assigned each child to one of the 6 conditions. True experiment. Step 3: When do you measure?

Just once, after treatment → posttest only Before AND after treatment → pretest-posttest Many times over time → time series

For this study: kids answered some demographic questions before viewing the video, but those don't count as a pretest because they're not measuring the same construct as the posttest. The DVs (perceived intent, recall, purchase intention) were measured only after the video. Posttest-only. Step 4: Where does it happen?

Lab/controlled → lab experiment Real-world → field experiment Studying something that already happened in the world → natural experiment

For this study: it's tricky because the standard categories don't fit cleanly. The study was conducted online, with kids on whatever device they had at home, possibly with parents nearby, possibly with siblings making noise. It's not a controlled lab environment, but it's not a traditional field experiment either. The authors flag this in their limitations: data collection was online, prohibiting verification of treatment fidelity. I'd call this an online or remote experiment — a hybrid that has some advantages for external validity (kids in real-world environments) but loses the controlled conditions of a lab. Step 5: Who experiences what?

Same people experience all conditions → within-subjects Different people in each condition → between-subjects (most common)

For this study: each kid saw only one combination of sponsorship cue and training. Between-subjects. Putting it together: This is a 3 × 2 factorial, true, posttest-only, online, between-subjects experiment. On the quiz, your answer should walk through each step and justify it with evidence from the article. The Experiment Types Overview infographic in Module 7 has the full taxonomy with definitions for each design type.

Here's a video about how causality was supported for each hypothesis/RQ in THIS study: https://youtu.be/LUbIPR9xBwM

Video on figuring out what type of experiment THIS study is: https://youtu.be/hetuKB9Uzp8

Video on figuring out the bigger goal in THIS study: https://youtu.be/EE-5Vc-U7dc

Here's an overview video for THIS study: https://youtu.be/yETbV8fVvm4

Video about THIS discussion section: https://youtu.be/Ldzj1L0TCCI

Video on figuring out what type of experiment THIS study is: https://youtu.be/hetuKB9Uzp8

Video on sampling, specific to THIS study: https://youtu.be/J4XljzuckHg

You may be familiar with these authors' universities, but there will be times when you're not - especially when reading research from other countries. When you can't easily evaluate an author's institutional affiliation because you're unfamiliar with the university, here are some strategies: * First, check whether the university has a dedicated research profile for the author - most universities worldwide maintain faculty pages in English, even if the institution's primary language is different. Look for their publication record, research focus, and degree. That is the most credible source. * Second, use Google Scholar to look up the author directly. Google Scholar works across languages and countries. You can see how many publications they have, how often they're cited, and whether they publish in journals you recognize as credible. An author with dozens of publications in peer-reviewed journals indexed in major databases is credible regardless of whether you've heard of their university. * Third, remember that credibility doesn't require name recognition. There are thousands of legitimate research universities worldwide that produce high-quality peer-reviewed scholarship. Not recognizing a university doesn't make it less credible - it just means you need to do a bit more digging. The same tools work everywhere: Google Scholar for the author, SCImago or Ulrich's for the journal, and the article's own reference list and citation count for the study itself. * What WOULD be a red flag: an author with no findable academic profile anywhere, a university that doesn't appear to exist or has no research output, or a journal that isn't indexed in any major database. Those are substantive credibility concerns - not just unfamiliarity.

Future research. Often comes directly from limitations. The authors suggest: One: examine non-health products. Two: vary video stimuli to look at affective reactions, not just cognitive. Three: compare across formats — online influencer versus TV commercial. Four: study reactions to known influencers, where parasocial relationships exist. Each of those addresses a specific limitation. That's the typical pattern.

RELATE TO PRIOR WORK: The authors invoke a theoretical explanation from prior work (Opree & Rozendaal, 2015; Rozendaal et al., 2009; Sagarin et al., 2002) to interpret their own finding. Kids who feel they "know" advertising might paradoxically be MORE susceptible because they don't deploy persuasion resistance — overconfidence in their own ad-detection skills makes them lower their guard. This explains why training had to do affective work, not just cognitive work, to be effective.

Author and journal credibility. At the end of the article, there's a "Notes on contributors" section. Sarah Vaala has a Ph.D. from the Annenberg School at the University of Pennsylvania — top-tier communication program — and she's an Associate Professor at High Point University. Her research focuses on persuasive messages, youth, and family decisions about media. Directly in her area of expertise. She's also affiliated with the Joan Ganz Cooney Center at Sesame Workshop, which is a research institute focused on children's media. The other authors are graduate-level — Francesca Mauceri completed a master's, Olivia Connelly a bachelor's, both at High Point.

Limitations: One: the sample isn't representative — recruited from an online panel, skews white, more educated, higher income. Two: 91.6% of the sample uses YouTube, so results may not apply to non-users. Three: data collection was online, so treatment fidelity wasn't verified. Four: only one set of stimuli was used. Five: the product was a health item, which differs from the toys, cosmetics, or food that dominate real youth-targeted unboxing. Six: ceiling effects — mean scores were high across measures. Seven: social desirability bias — kids may have answered the way they thought researchers wanted. Eight: dental health information may have been pre-known by older tweens.

When researchers run a factorial design (like this 3×2 study), they don't just check whether each IV has its own effect — they also check whether the IVs interact with each other. An interaction effect tells you that the impact of one IV depends on the level of the other IV. Let's make this concrete. Imagine an experiment testing whether caffeine improves test performance, with two IVs: caffeine (caffeine vs. no caffeine) and time of day (morning vs. evening). Three different patterns could emerge:

Main effect of caffeine only: caffeine helps everyone equally, no matter what time of day. The effect of caffeine doesn't depend on time of day. Main effect of time only: people perform better in the morning regardless of caffeine. The effect of time doesn't depend on caffeine. Interaction: caffeine helps in the morning but doesn't help in the evening. The effect of caffeine depends on the time of day. That's an interaction.

Interactions are often the most interesting findings in a factorial study. They show that the world is more complex than a single IV's effect — that effects depend on context. In this study, the authors are looking for an interaction between sponsorship cue and training video. Their hypothesis is essentially: training video changes how kids respond to sponsorship cues. They use Wald tests to check whether the relationship between perceived intent and the outcome is different for participants who saw the training video versus those who didn't. When the Wald test is significant (p < .05), there's an interaction — the effect of one IV depends on the other. The actual findings include several significant interactions: training moderates the relationship between perceived informative intent and purchase intention, and training moderates the relationship between perceived informative intent and health recall. In both cases, training changes how perception predicts the outcome. When you read an experimental study on the quiz, look for:

Main effects — does each IV have its own effect on the DV? Interaction effects — do the IVs jointly influence the DV in a way that's more than the sum of their separate effects?

A significant interaction is often more theoretically interesting than the main effects, because it reveals a contingency — when the effect happens, when it doesn't. Extra video on factorial designs and interaction effects: https://www.youtube.com/watch?v=2wZAAQ6OdFw

This is ALMOST like a cover story, or at least does SOMETHING to reduce demand characteristics.

Pause here. Before you analyze this hypothesis, you need to understand a concept that the Module 7 lecture didn't go deep on: in some experiments, an independent variable is measured rather than manipulated. This study is one of those cases — and it changes everything about how we evaluate causality. The Module 7 lecture defines an IV as "the thing that is manipulated or somehow introduced into the setting by the experimenter." That's the standard case, and it's true for the sponsorship cue and training video in this study — those are manipulated IVs. The researchers control them by random assignment. But this study also has IVs that are measured, not manipulated. Perceived selling intent and perceived informative intent are measured. The researchers didn't control these — they didn't make some kids perceive selling intent and other kids not. They asked all kids how much selling intent they perceived, and recorded what each kid reported. Why does a researcher use a measured IV in an experiment? Sometimes the construct can't be ethically or practically manipulated. You can't randomly assign kids to "perceive a video as sponsored" — perception is something that happens in their head based on what they see. The researchers can manipulate the cue (sponsored vs. non-sponsored video) and measure the resulting perception, but they can't directly control the perception itself. Why does this matter? Because random assignment is what gives experiments their causal power. When an IV is randomly assigned, you can rule out confounds — pre-existing differences between groups should be roughly equal. When an IV is measured, no random assignment happened, so confounds remain a real concern. Practical implications:

For manipulated IVs (sponsorship cue, training video): random assignment supports strong claims about temporal order and non-spuriousness. Causality is well-supported by the design. For measured IVs (perceived selling intent, perceived informative intent): no random assignment. Temporal order is fuzzy because perception and outcomes are measured close together. Non-spuriousness is weak because confounds (advertising literacy, prior product interest, working memory) could explain both perception and outcome.

So when you analyze each hypothesis in this study, the FIRST question to ask is: is the IV manipulated or measured? That answer shapes how strongly you can claim causality. Track the variables as you read:

Manipulated: sponsorship cue (3 levels), training video (2 levels) Measured: perceived selling intent, perceived informative intent

Any hypothesis with a measured IV will have weaker causal claims than a hypothesis with a manipulated IV — even though the overall study uses random assignment to conditions.

Since this study DOESN'T have manipulation checks, let me show you what a manipulation check actually looks like in the wild, so you can recognize one if you see it on the quiz. Imagine a study testing whether watching a sad video makes people more likely to donate to charity. The IV is the video (sad vs. neutral), and the DV is donation behavior. A manipulation check would ask, right after the video: "How would you describe your current mood?" If participants in the sad video condition rated their mood as significantly more sad than participants in the neutral condition, the manipulation worked. If both groups reported similar moods, the manipulation failed — and any null result on donation behavior couldn't tell us whether sadness doesn't affect donations or whether the video just didn't make people sad. Or imagine a study where some participants watch a video featuring a "familiar" character and others watch one featuring an "unfamiliar" character. A real manipulation check would ask each participant if they recognized the character or could name them — if kids in the "familiar" condition all named the character correctly and kids in the "unfamiliar" condition couldn't, the familiarity manipulation worked. The pattern is always: a quick, direct question about the manipulation itself, asked separately from the dependent variable. It's not the same thing as the DV. It's a check on whether the IV did what it was supposed to do. When you encounter a different study on the quiz, look for manipulation checks in the methods or results sections. They might appear as a sentence like "Participants in Condition A reported significantly higher [thing manipulated] than participants in Condition B, t(X) = Y, p < .05." That's the manipulation working as intended.

FUTURE RESEARCH: Real influencer testing. The unknown actor used in this study may have produced different reactions than a real influencer kids actually follow. Future research should test reactions to known influencers, where parasocial relationships are already established.

Extra video on cover stories: https://www.youtube.com/watch?v=DHfvtcMvKeA&feature=youtu.be

Extra video on manipulation checks: https://www.youtube.com/watch?v=0mr2K9Pji7k

RELATE TO PRIOR WORK: The authors connect their finding (that informative intent was perceived universally, even in sponsored conditions) to research by Naderer et al. (2021) on parasocial perceptions. The unknown peer-aged actor in this study may have triggered the same authenticity heuristic real influencers do — kids perceive sameness as honesty. This both validates their finding and points to where future research should go.

LIMITATION: Pre-existing knowledge. Older participants may have already known the dental health information from school, which would inflate health recall scores independent of what they learned from the video. The recall measure can't distinguish "learned from this video" from "already knew."

Extra video on demand characteristics: https://www.youtube.com/watch?v=dSFIAoTKb0o&feature=youtu.be

RELATE TO PRIOR RESEARCH: The authors connect their finding (that perceived selling intent positively predicts purchase intention without training) to a body of prior research showing the same surprising pattern in adolescents (Harms et al., 2022; Opree & Rozendaal, 2015; Vanwesenbeeck et al., 2016a, 2016b). This is counterintuitive — you'd expect detecting an ad to make kids LESS interested, but the literature consistently shows the opposite for some kids. Their finding fits this pattern.

FUTURE RESEARCH: Affective reactions. The authors note their study focused mostly on cognitive reactions (recall, perceived intent). Future research should also measure affective reactions — emotional responses, trust, liking — which might mediate or moderate purchase intention in ways the current study didn't capture.

LIMITATION: Social desirability bias. Kids might have answered the way they thought the researchers (or their parents nearby) wanted them to answer — saying they'd ask for the product when they wouldn't, or saying they perceived selling intent because that seems like the "smart" answer.

FUTURE RESEARCH: Product category replication. The authors call for replication using toys, cosmetics, food, and other youth-targeted product categories rather than just health products. Different product types might trigger different schemas and different levels of skepticism.

FUTURE RESEARCH: Cross-format comparison. The authors suggest examining the same kind of content across different formats — for example, an unboxing video vs. a child actor in a traditional TV commercial — to see whether tweens have different mental schemas for different ad formats.

LIMITATION: Product type. The water flosser is a health product, while most real unboxing content features toys, cosmetics, or food. Tweens might respond differently — possibly with less skepticism — to the product categories they actually encounter on YouTube.

LIMITATION: Stimulus generalizability. The researchers tested ONE specific set of videos. Findings might not extend to videos with different hosts, different production styles, different products, or different platforms. Replication with varied stimuli is needed.

LIMITATION: Ceiling effects. Most participants scored near the top of the scales for several measures, which limits how much variation the analysis can detect. Real differences might exist that the data can't reveal because everyone is clustered at the high end.

LIMITATION: Sample representativeness. The authors recruited from Dynata's online panel, which skews toward more-educated, higher-income, mostly White households. Findings may not generalize to U.S. tweens from less represented demographic groups.

LIMITATION:Treatment fidelity. The researchers couldn't directly verify that participants actually watched the videos as intended. Kids could have skipped sections, been distracted, or had a parent help them — and there's no way to tell from online data collection.

LIMITATION: YouTube familiarity. Almost all participants are existing YouTube users, so they're already familiar with unboxing-style content. Tweens who don't use YouTube — possibly the most vulnerable population because they're newer to this content format — aren't represented.

Tables 4 and 5 are dense with statistics. Let me break down what you're looking at, because you'll see versions of these on the quiz. The b value (unstandardized regression coefficient). This tells you how much the dependent variable changes for each one-unit increase in the independent variable. So if b = 0.59 for perceived informative intent → purchase intention, that means: for each 1-point increase in perceived informative intent, purchase intention increases by 0.59 points (on its own scale). The sign of b matters too:

Positive b: the IV and DV move in the same direction — when one goes up, the other goes up. Negative b: the IV and DV move in opposite directions — when one goes up, the other goes down.

So in this study, b = 0.59 (positive) means more perceived informative intent leads to more purchase intention. If you see b = −0.25, it would mean more of one variable leads to less of the other. The confidence interval [CI b] in brackets after the b value. This is the range of plausible values for b. If the confidence interval includes zero, the relationship isn't statistically significant (you can't rule out the possibility that the true effect is zero). If the interval doesn't include zero, the relationship is statistically significant. Example from the article: b = 0.59 [0.38, 0.79] means: the best estimate for the effect is 0.59, and the true effect is plausibly anywhere between 0.38 and 0.79. Since this interval doesn't include zero, the effect is statistically significant. Compare to b = 0.10 [−0.03, 0.22] — this interval DOES include zero, so the effect is NOT statistically significant. Effect sizes more broadly. A b value tells you the size of an effect, but other studies report different metrics for effect size:

Cohen's d — used for comparing two means; .2 is small, .5 is medium, .8 is large. Pearson's r — correlation coefficient; .1 is small, .3 is medium, .5 is large. Partial eta squared (η²p) — used in ANOVA; .01 is small, .06 is medium, .14 is large. You'll see this metric if a different study uses ANOVA instead of regression. Odds ratios, R², and others — different statistical methods produce different effect size metrics.

Why effect sizes matter beyond p-values. A statistically significant result with a tiny effect size might be technically real but practically meaningless — especially with large samples, where even trivially small effects can reach statistical significance. Always look at effect size alongside p-values to assess whether a finding is practically important.

Perceived informative intent (linked to Variable 6). From schema theory: informative intent is recognized when content is identified as objective/educational, activating an educational schema that directs attention toward learning rather than consumer response.

No conceptual definition is provided in the article. The authors describe operationalization but never define this theoretically.

No conceptual definition is provided in the article. The authors describe operationalization but never define this theoretically.

Perceived selling intent (linked to Variable 7). From the Persuasion Knowledge Model: selling intent is recognized when consumers identify that a message is trying to persuade them to buy something, prompting cognitive coping responses.

Training video / advertising literacy (linked to Variable 2: Training video / no training video). The authors define advertising literacy as the cognitive skills children use to filter and critically evaluate the intent behind advertising content. The training video manipulation is meant to teach or activate exactly these skills.

Compensation. Participants are paid (in points convertible to money or prizes) for participating. Compensation is ethically standard but worth noting because it can create subtle pressure to participate — especially for people who depend on panel earnings as supplementary income.

Right to refuse and right to withdraw. Three children declined to participate even after their parents had consented — their decisions were respected. Thirteen others quit partway through and weren't forced to continue. Both reflect the ethical principle that participation must be voluntary and ongoing, not just at the start.

Sponsored video / unboxing host (linked to Variable 1: Non-sponsored / Sponsored / Unaddressed sponsored conditions). The authors define an "influencer" — and by extension a sponsored video — as content where the host has a paid or otherwise compensated relationship with the manufacturer.

The study was reviewed and approved by an Institutional Review Board before data collection — required for any university-affiliated research with human participants. Research involving minors typically receives heightened IRB scrutiny because children are a "special population" with reduced capacity for consent.

Key term to understand for this study. Native advertising is advertising that's designed to look like the regular content around it, so consumers don't immediately recognize it as an ad. Examples in different formats:

In a magazine: a "sponsored article" that reads like editorial content but is actually paid promotion. On Instagram: a post by an influencer wearing a brand's clothes that looks like personal content but is sponsored. On YouTube: an unboxing video where the host reviews a product they were paid to feature, presented in their normal vlog style.

The point of native advertising is to evade the consumer's "ad detection" defenses. When you see a TV commercial, you know it's an ad — there's a clear break, a jingle, a brand logo, a 30-second window. Your brain switches into "this is trying to sell me something" mode. With native advertising, those visual and structural cues are missing, so your defenses don't activate. This is exactly what the FTC sponsorship disclosure rules are trying to address. By forcing a clear "sponsored" label on native ads, the rules give consumers the cue they need to recognize the content as advertising and engage their critical processing. Why does this matter for kids? Their advertising literacy is still developing. Even when kids can spot traditional ads on TV, they often DON'T spot native ads embedded in entertainment content like unboxing videos. That's the puzzle this study is trying to address.

This study uses TWO theoretical frameworks at once: Schema theory and the Persuasion Knowledge Model (PKM). That's worth understanding because theoretical frameworks do a lot of work in research articles, and being able to spot them is a meta-skill for reading any empirical paper. What theoretical frameworks do:

They explain WHY the researchers expect specific relationships to exist. Without a framework, hypotheses are just guesses; with one, they're predictions grounded in existing theory. They organize the study's structure (which variables to include, what to measure, what to compare). They give meaning to the results. A finding "matters" because it supports, contradicts, or refines an existing theoretical understanding.

What's happening in THIS study:

Schema theory says people organize knowledge into mental schemas that direct attention and processing. The authors use schema theory to predict that kids who detect a SELLING intent will engage their advertising schema (focusing on product details), while kids who detect an INFORMATIVE intent will engage their educational schema (focusing on health info). PKM says when consumers detect persuasive intent, they engage cognitive coping strategies (often becoming more critical, more resistant). The authors use PKM to predict that kids who detect selling intent might develop defensive processing.

Notice these two frameworks make slightly different predictions. Schema theory predicts MORE recall of relevant info; PKM predicts MORE critical processing. The authors test both possibilities, which is part of what makes RQ1 interesting (they're asking, essentially, "which framework wins out?"). When you read any empirical study, look for the theoretical framework(s) early on. They tell you what the authors expect and why. Findings that match the framework's predictions are theoretical wins; findings that don't push the framework to evolve.

Informed consent for parents. Parents read the study description and explicitly consented before any data collection began. This is the standard ethical baseline for adult participation.

This sentence reflects an important ethical distinction in research with minors: parents give consent, children give assent. Consent is legally meaningful authorization that the participant fully understands what they're agreeing to. People under 18 are not considered legally capable of giving consent on their own behalf, so a parent or guardian gives it for them. Assent is the child's own agreement to participate, given in age-appropriate language they can understand. It's not legally binding the way consent is, but research ethics requires it because children deserve agency over their own participation. A parent's consent doesn't override a child's refusal — if the child says no, even after the parent says yes, the child doesn't participate. This study did this correctly:

Parent reads the consent form and consents on the child's behalf. Parent brings child to the device. Child reads (or hears) the assent language in age-appropriate terms. Child decides for themselves whether to participate.

The authors note three children declined to participate at the assent stage — their refusals were honored. That's research ethics in action. Without an assent process, those kids' parents could have effectively forced them into the study. This is one of the implicit ethical considerations you should be taking note of. The authors don't shout about it, but the practice is significant.

Worth pausing on this. Purchase intention — arguably the most important DV in the study, since it's about real-world consumer behavior — was measured with a SINGLE item: "Will you ask your parents to buy a [company] water flosser?" answered on a 5-point scale. Single-item measurement creates several measurement issues: Reliability is unmeasurable. You can't compute Cronbach's alpha with one item — there's nothing to compare it to internally. So we have no statistical evidence that this measure is reliable. Validity rests entirely on this one question. If the question doesn't quite capture what "purchase intention" really means — maybe kids interpret "ask my parents" as different from "want this product" — there's no second item to triangulate against. Variability is constrained. With only 5 response options and 251 kids, you'll get clusters at certain values, which limits your ability to detect subtle effects. Researchers sometimes use single-item measures because the construct is simple, the survey needs to be short (especially with kids), or face validity is high. But it's a real measurement weakness, and worth flagging when you assess the study's quality. Compare this to perceived informative intent (3 items, α = .82) and perceived selling intent (2 items, r = .53). Those are more reliable, even though selling intent is also fairly thin.

Important note about this study: it does NOT include manipulation checks or attention checks. These are different things, and the absence of both is worth noticing. A manipulation check verifies that the IV manipulation actually worked — that participants noticed the manipulation and responded to it as the researchers intended. For this study, manipulation checks would have asked things like:

Did the kid notice the sponsorship disclosure in the sponsored condition? Did kids correctly understand that the school report video was non-commercial? Did kids actually absorb the content of the training video?

Without manipulation checks, we don't know whether the manipulations worked. If results are null (like H4, H6b), is that because there's truly no effect, or because the manipulation never landed? An attention check verifies that participants were paying attention generally — not that the manipulation worked, but that they were focused enough on the study for any of their responses to be meaningful. Common attention checks include items like "select 'strongly disagree' to show you're reading carefully" embedded in surveys. The authors actually flag the absence of attention checks in their limitations: they note "Recruitment and data collection were also conducted online, prohibiting verification of treatment fidelity." Treatment fidelity = whether the participants actually received the treatment as intended. Be careful not to confuse perceived selling intent and perceived informative intent with manipulation checks. They look similar (asking kids what they thought of the video), but in this study they're treated as mediators and DVs, not as checks. A real manipulation check would ask something like "Was this video sponsored, yes or no?" — a direct test of whether the manipulation registered. Extra video on manipulation checks: https://youtu.be/0mr2K9Pji7k

Take a minute with this figure — once you can read these diagrams, you can identify the variable structure of any study in seconds. Conceptual models like this one have a standard grammar:

Boxes are constructs (variables, conditions, outcomes). Arrows show predicted relationships, usually IV → DV (cause → effect). Position shows the role: leftmost are IVs/conditions, rightmost are outcomes/DVs, anything in the middle is a mediator (something on the path between IV and DV). A box with an arrow pointing AT another arrow (like Pre-Roll Training pointing at the path between Perceived Intention and Outcomes) is a moderator. It changes the strength or direction of a relationship without sitting on the causal path itself.

Read this specific figure: experimental conditions (sponsored, non-sponsored, unaddressed) → perceived intention (informative or selling) → outcomes (recall, purchase intention). And training video moderates the second arrow. So the variable structure of the whole study is:

IV (manipulated): sponsorship cue (3 levels) Mediators (measured): perceived informative intent, perceived selling intent Moderator (manipulated): training video (2 levels) DVs (measured): product recall, health recall, purchase intention

When you encounter a complex study like this one, look for the conceptual model first. It tells you what role each variable plays before you wade into the methods section.

RQ2b (perceived intent × training → purchase intention): Association: Supported. Perceived informative intent predicted purchase intention in BOTH conditions, but more strongly with training (b = 0.95) than without (b = 0.59, Wald test p < .05). Perceived selling intent predicted purchase intention only WITHOUT training (b = 0.29) — training eliminated this relationship.

Temporal order: Partial. Same pattern as RQ2a — moderator's role is well-ordered (training came first), but the IV (perceived intent) is measured, blurring time order for the IV → DV link.

Non-spuriousness: Partial. Training is randomly assigned (supports moderator inference), but perceived intent isn't (leaves the door open for confounds like product interest, advertising familiarity, general skepticism).

Verdict: The moderating role of training is causally supported. The underlying IV → DV relationship between perceived intent and purchase intention is correlational.

Pause here. This hypothesis is about a relationship between two variables — perceived selling intent (IV) and product recall (DV) — and it's worth understanding upfront that perceived selling intent is a measured variable, not a manipulated one. What's the difference, and why does it matter? A manipulated variable is one the researcher controls directly. They assign participants to different levels (training video vs. no training video, sponsored vs. non-sponsored) and watch what happens. Random assignment makes manipulated variables powerful for causal inference. A measured variable is one the researcher just observes — they survey participants and record what they say. Perceived selling intent is measured: kids watched a video, then answered survey questions about whether they thought the host was being paid. The researchers didn't make some kids perceive selling intent and other kids not perceive it. Why this matters: when an IV is measured (not manipulated), causality claims get much weaker. You can't establish strong temporal order (because perception and outcome are measured close together). You can't rule out confounds via random assignment (because you didn't assign anyone to anything). Pre-existing differences between high-perceivers and low-perceivers might explain BOTH their perception and their recall. When you read the hypotheses and results, keep track of which variables are manipulated and which are measured. The training video and sponsorship cue are manipulated. Perceived intent (both kinds) is measured. That distinction will shape every causality answer you give.

RQ2a (perceived intent × training → recall): Association: Supported. Significant moderation by training video on the relationship between perceived informative intent and health recall (Wald test p < .05). With training, perceived informative intent predicted health recall (b = 0.21); without training, it didn't.

Temporal order: Moderate. The MODERATOR (training) was manipulated and came first by design. But the IV (perceived intent) was measured, not manipulated, and assessed alongside the DV. So time order is fuzzy for the IV → DV link, even though it's clean for the moderator's role.

Non-spuriousness: Partial. Training was randomly assigned, supporting strong inference about the moderator's role. But perceived intent wasn't, so third-variable explanations remain possible for the core IV → DV link.

Verdict: The moderating role of training is causally supported. The underlying IV → DV relationship is correlational, not causal.

H4 (non-sponsored → perceived informative intent): Association: NOT supported. Tweens perceived similar informative intent for sponsored and non-sponsored videos, regardless of training condition. [FYI - even if association is unsupported, you still need to go through the rest of the steps!]

Temporal order: Strong. Video condition was manipulated and presented before measurement.

Non-spuriousness: Strong. Random assignment controls for pre-existing differences.

Verdict: Causality cannot be claimed because there's no association to begin with. Despite a strong experimental design, the predicted relationship simply didn't appear in the data. Good design + no effect = no causal claim.

H6b (training video → detection of informative intent in non-sponsored content): Association: NOT supported. No significant difference in perceived informative intent between training and no-training groups, in any condition. [FYI - even if it was unsupported, still walk through the rest of the steps.]

Temporal order: Strong (same reasoning as H6a — training came first by design).

Non-spuriousness: Strong (random assignment).

Verdict: Causality cannot be claimed because there's no association. Despite a strong experimental design, the predicted relationship didn't appear.

H6a (training video → detection of selling intent in sponsored content): Association: Partial. The training × sponsorship moderation wasn't significant overall, but within the sponsored condition specifically, training did increase perceived selling intent (b = 0.50, p < .01).

Temporal order: Strong. Training video came first, sponsored video second, perception measure third.

Non-spuriousness: Strong. Random assignment to both training/no-training and to sponsorship condition. Standardized stimulus exposure.

Verdict: Causality conditionally supported within the sponsored condition. Strong experimental design supports the inference.

RQ2a: Perceived video intent — both informative and selling (measured IV) → training video (manipulated moderator) → recall of information, both product and health (DV)

H2 (perceived informative intent → educational recall): Association: Partial — perceived informative intent predicted greater health recall, but ONLY among tweens who viewed the training video (b = 0.21, p < .05). Without training, no significant relationship.

Temporal order: Moderate. Perception was measured before recall, so sequence exists. But the IV is measured (not manipulated), which weakens the strength of any temporal claim.

Non-spuriousness: Weak. Same problem as H1 — perceived informative intent wasn't randomly assigned, so confounds (motivation, prior interest in health, engagement with the video) could shape both perception and recall.

Verdict: Causality only weakly supported, and only conditional on training. The schema activation mechanism remains theoretical — it's not directly tested.

H2: Perceived informative intent (measured IV) → [theoretical mediator: educational schema activation] → health/educational recall (DV)

H3 (sponsored video w/ disclosure → perceived selling intent): Association: Partial — tweens perceived greater selling intent in the sponsored vs. non-sponsored condition, but ONLY when they had first viewed the training video (b = 0.50, p < .01). Without training, no significant difference.

Temporal order: Strong. Sponsorship was a manipulated IV, randomly assigned, presented BEFORE perceived selling intent was measured.

Non-spuriousness: Strong. Random assignment to sponsorship condition controls for pre-existing differences. All participants saw the same video format with only the sponsorship disclosure varying.

Verdict: Causality conditionally supported — the sponsorship disclosure DOES affect perceived selling intent, but only in the presence of advertising literacy training.

H1 (perceived selling intent → product recall): Association: Partial — there IS a statistical relationship, but it runs OPPOSITE to predicted. Higher perceived selling intent was associated with LOWER product recall, especially after training (p < .05). This contradicts H1 but is consistent with the Persuasion Knowledge Model (defensive processing).

Temporal order: Weak. Perceived selling intent was MEASURED, not manipulated. Both perception and recall are responses to the same video, so we can't strictly establish that perception came first in time.

Non-spuriousness: Weak. Because the core IV wasn't randomly assigned, third variables (advertising literacy, skepticism, working memory capacity) could explain both perception and recall.

Verdict: Causality NOT established. The key IV is measured, not manipulated, and the association ran opposite to predicted.

A few tips for working through these variable questions: On level of measurement: Don't just assume. Read the actual response options. For example, an income question that asks people to pick from ranges (under $50K, $50–99K, etc.) is ordinal, not interval/ratio, even though income itself is a continuous construct. Watch for this — students lose points on it constantly. On variable role (IV / DV / mediator / moderator / control): A variable can play DIFFERENT roles in different hypotheses. Don't pick one and call it good. Walk through each hypothesis (H1, H2, H3...) and ask: "in this specific hypothesis, what role does this variable play?" List them all, like "DV in H1, RQ2a; IV in H6b." On measurement validity/reliability: Look for Cronbach's alpha (α) for multi-item scales, and look for any mention of validation studies or pilot testing. If the authors are silent on this, note that — silence isn't always a problem (some scales are well-established), but it's worth flagging. On conceptual definition: This is the abstract, theoretical definition of the construct — usually in the introduction or literature review. It's NOT the operationalization (the specific items used to measure it). Look for a sentence that defines what the construct is, not how it's measured. Sometimes you'll need to read carefully — conceptual definitions can be embedded in the middle of paragraphs, not always called out explicitly.

RQ2b: Perceived video intent — both informative and selling (measured IV) → training video (manipulated moderator) → purchase intention (DV)

Heads up before you tackle this one: the moderator in this hypothesis is theoretical, not measured. The authors don't have a direct survey item for "schema activation" — it's an internal cognitive process that's inferred from the pattern of results, not assessed directly. So when you identify the moderator here, you're naming the theoretical mechanism the authors propose: schema activation (advertising schema for H1, educational schema for H2). The hypothesis says perceiving selling/informative intent should trigger the schema, which then directs attention to certain kinds of information, leading to better recall of that information. In the path analysis the authors actually run, schema activation isn't a separate variable. The study tests the IV → DV link directly and uses schema theory to explain WHY that link exists. So the moderator is theoretical, not statistical. This is a good lesson in how published research works. Authors often name theoretical moderators or mediators in their hypotheses but only test some of them directly with measured variables. Read carefully to figure out which constructs are measured vs. which are described conceptually as the "reason why."

Students sometimes confuse the "bigger research question" with the specific hypotheses. They're not the same. The specific hypotheses are narrow, testable predictions: "tweens will perceive greater selling intent in a sponsored unboxing video compared to a non-sponsored video." That's H3. It's specific. It involves named variables. It can be supported or unsupported by data. The bigger research question is the larger problem the study is trying to address — the reason any of those specific hypotheses are worth testing in the first place. It's the answer to "why bother?" It's what you'd tell your grandma if she asked what the study is about. For this study, the bigger picture might be something like: Kids spend tons of time watching online videos, and a lot of that content is secretly trying to sell them stuff. We know how kids handle traditional ads, but unboxing videos and influencer content are a newer beast — they don't look like ads, even when they are. Tweens are at a developmental moment where they're capable of detecting commercial intent but might not do it automatically. So can we figure out (1) whether kids perceive sponsored content differently than objective content, and (2) whether brief media literacy training helps them apply skills they already have? Notice what that does: it zooms out. It explains what's at stake (kids being manipulated by stealth advertising). It explains why this age group matters (developmental window). It explains why this study contributes (existing research focuses on traditional ads, not native digital content). And it doesn't get into the specific 6 hypotheses. A good "bigger research question" answer should be a paragraph or so. Write it in your own words, not paraphrasing the abstract. Imagine explaining the point of the study to someone who's never read it.

H1: Perceived selling intent (measured IV) → [theoretical mediator: advertising schema activation] → product recall (DV)

Once you've got the conceptual difference between mediators and moderators down, the next skill is spotting them in published studies — which is harder than it sounds, because authors don't always label their variables that way explicitly. Here are the textual cues to watch for: Moderator language: "the relationship between X and Y depends on Z," "Z moderated the effect of X on Y," "the effect of X on Y was stronger/weaker when…," "Z × X interaction," "for participants high in Z, the effect was…" Statistical tests for moderation usually involve interaction terms in regression or ANOVA. Mediator language: "X influences Y through Z," "Z explains the relationship between X and Y," "X has an indirect effect on Y via Z," "Z is the mechanism by which X affects Y." Statistical tests for mediation include path analysis (which this study uses), Baron and Kenny's steps, or bootstrapped indirect effects. In this specific study, the authors are very clear about the training video as a moderator — they say so explicitly in the conceptual model in Figure 1, and they test it using "Wald statistics tested whether model paths varied significantly between children who saw the advertising training video and those who did not." But they're also using perceived informative intent and perceived selling intent as mediators — variables that sit between the experimental condition (IV) and the outcomes (recall, purchase intention). The path analysis structure tells you this: the experimental condition affects perceived intent, which then affects the DVs. Perceived intent is the mechanism the authors think explains why sponsorship cues affect kids' responses. Notice this study uses both at once. That's common in published research — moderators and mediators do different work in the same model. Extra video on finding moderators and mediators in published studies: https://youtu.be/nNcrmLLR_Rc

Heads up before you tackle this one: the moderator in this hypothesis is theoretical, not measured. The authors don't have a direct survey item for "schema activation" — it's an internal cognitive process that's inferred from the pattern of results, not assessed directly. So when you identify the moderator here, you're naming the theoretical mechanism the authors propose: schema activation (advertising schema for H1, educational schema for H2). The hypothesis says perceiving selling/informative intent should trigger the schema, which then directs attention to certain kinds of information, leading to better recall of that information. In the path analysis the authors actually run, schema activation isn't a separate variable. The study tests the IV → DV link directly and uses schema theory to explain WHY that link exists. So the moderator is theoretical, not statistical. This is a good lesson in how published research works. Authors often name theoretical moderators or mediators in their hypotheses but only test some of them directly with measured variables. Read carefully to figure out which constructs are measured vs. which are described conceptually as the "reason why."

These are in fact variables! Race/ethnicity, household income, and parental education are demographic variables collected from parents at the start of the survey. To figure out how they're being used, ask: does this variable show up in any of the hypotheses or RQs? Look back at H1–H6, RQ1, RQ2. None of them mention race/ethnicity. None mention income. None mention parental education. So these aren't IVs or DVs in any of the tested relationships. That tells you these are control variables (also called covariates). Researchers collect them for two main reasons:

To describe the sample. That's what Table 1 is doing — telling you who actually participated so you can judge external validity. To statistically adjust for them in the analyses. This is when researchers want to make sure the effects they're seeing aren't driven by demographic differences across conditions. (In a true experiment with random assignment, demographics should be roughly equal across conditions anyway — but checking and statistically controlling is good practice.)

For each of these demographic variables, you'll want to identify:

Operationalization: what specific question was asked, and what response options were given? (The article is sparse on this — for some demographics, you may have to infer from Table 1 what the response options must have been.) Level of measurement: race/ethnicity is nominal (categories with no order). Education is typically ordinal (less than bachelor's < bachelor's < graduate). Income — be careful here. If they reported it as ranges (under $50K, $50–99K, etc.), that's actually ordinal, not interval, even though the underlying construct is continuous. Use: control variable, not part of any hypothesis.

Extra video on finding IVs and DVs (and what's not an IV or DV): https://youtu.be/SlljkpUY4J4

Quick refresher on what a manipulation check is and why it matters for this study. A manipulation check is a measurement researchers add to verify that their IV manipulation actually worked the way they intended — that participants noticed it, understood it, or were affected by it. It's not the dependent variable. It's a check on the integrity of the experimental setup itself. For example, in a study manipulating "fear appeals" in health messages, the researcher would want to verify that participants in the high-fear condition actually felt more afraid than participants in the low-fear condition. If the manipulation check shows no difference in self-reported fear between groups, the manipulation failed — and any downstream findings on the DV become hard to interpret. Maybe there's no effect because the IV doesn't matter. Or maybe there's no effect because the manipulation never actually worked. For each of the experimental conditions in this study (non-sponsored, sponsored, sponsorship unaddressed, training video), ask yourself:

Did the authors verify that participants noticed the manipulation? If yes, what did they ask, and what did they find? If no, what could they have asked? And how does the absence of a check affect your confidence in the findings?

Be careful here — perceived selling intent and perceived informative intent are dependent variables in this study (and mediators in the path analysis), not manipulation checks, even though they look similar to what a manipulation check might measure. A true manipulation check would ask something like "was this video sponsored?" — a direct test of whether kids registered the sponsorship cue itself. Extra video on manipulation checks: https://youtu.be/0mr2K9Pji7k

This is the authors flagging an external validity concern, and it's worth pausing on because the trade-off it represents is one of the most important tensions in experimental research. Internal validity is about whether you can be confident the IV actually caused changes in the DV — whether you've ruled out alternative explanations. Tightly controlled experiments tend to have strong internal validity because you can rule out lots of confounds. External validity is about whether your findings generalize beyond the specific sample, setting, time, and stimuli of your study — whether what you found would hold up with other people, in other places, with other materials. These two often pull in opposite directions. The more controlled your experimental setting (which boosts internal validity), the less it resembles the real world (which hurts external validity). The more "ecologically valid" your setup (real environments, real materials, real behaviors), the harder it gets to rule out confounds. This study leans toward internal validity. The authors created controlled video stimuli, used random assignment, and standardized exposure through Qualtrics. That's good for causal inference. But it costs them on external validity:

The sample skews toward higher-income, more-educated, mostly White families on an online research panel — not the full diversity of U.S. tweens. The stimuli were custom-made videos with an unknown actor — not real influencer content with hosts kids actually follow. The product was a water flosser — not the toys, cosmetics, or food products that dominate real youth-targeted unboxing content. The viewing happened in a survey environment with a parent nearby — not a kid alone on YouTube doomscrolling at 9pm.

None of this means the findings are wrong. It means we should be cautious about generalizing them. The next step in this research program would be replication with more diverse samples, real influencer content, and naturalistic viewing contexts. Extra video on internal vs. external validity: https://youtu.be/ehq62uzVAzM

Author credibility - remember that we Google authors and look for specific credibility markers. These are all indicators of credibility. When evaluating author credibility, look for: institutional affiliation, degree and where it's from, research focus alignment with the study topic, and publication record. You could also look them up on Google Scholar to see their other publications and citation counts - but again, on a quiz you'd just need to know WHERE you'd look and what you'd look for.

Let's unpack this notation, because you'll see it constantly in experimental research and it's worth being able to read at a glance. When researchers write "3 × 2," they're describing a factorial design — an experiment that manipulates more than one independent variable at the same time. The numbers tell you the levels:

The first IV (sponsorship cue) has 3 levels: sponsored, non-sponsored, sponsorship unaddressed. The second IV (advertising training) has 2 levels: training, no training. Multiply those together and you get 6 total conditions. Each tween is randomly assigned to exactly one of those 6 cells.

This is between-subjects, which means each kid only sees one combination — not all six. That's why factorial studies often need larger samples than single-IV experiments: you're filling 6 cells, not 2. Here's why a factorial design is more powerful than just running two separate experiments. It lets you test for interaction effects — does the effect of one IV depend on the level of the other? That's exactly what the authors are after here. They don't just want to know "does sponsorship disclosure matter?" and "does training matter?" separately. They want to know whether training changes how kids respond to the sponsorship cue. A factorial design is the only way to answer that question in a single study. You'll see this notation everywhere: a 2 × 2 has 4 conditions, a 2 × 3 has 6, a 3 × 4 has 12, and a 2 × 2 × 2 (three IVs) has 8. The pattern always holds. Extra video on factorial design: https://www.youtube.com/watch?v=r0tn9E0WPks. The Identifying Experiment Types infographic in Module 7 also walks through how to count factorial conditions.

Two related concepts here that students often blur together: cover stories and demand characteristics. A cover story is what the researchers tell participants the study is about — and it's often not the real research question. Why? Because if you tell a tween "we're studying whether you can detect sponsored content," they're going to perform — they'll scan for sponsorship cues much more carefully than they would normally, and you'll get inflated detection rates that don't reflect real-world behavior. Demand characteristics are the problem cover stories try to solve. They're cues in the experimental setup that tip participants off to what the researcher expects, leading participants to (consciously or unconsciously) provide responses that match those expectations rather than their genuine reactions. Now look closely at this study. Ask yourself:

What did the kids think the study was about? (Look at the assent language and the procedures.) Were the videos presented as real YouTube content? (Yes — the authors note they streamed videos "from a private YouTube channel, in order to boost children's beliefs that these were real videos on the YouTube platform.") That framing is a deceptive element built into the stimulus, not just the study description. Did the authors explicitly mention using a cover story? Or were they relatively transparent about the topic?

This study sits in an interesting middle ground: there's no elaborate cover story about the study purpose, but there is mild deception about the authenticity of the video stimuli. Could the authors have run this study without any deception? What would that have cost in terms of internal validity? Extra video on cover stories: https://youtu.be/DHfvtcMvKeA Extra video on demand characteristics: https://youtu.be/dSFIAoTKb0o

Extra video on finding limitations: https://www.youtube.com/watch?v=A4CzgLntcOc

Extra Video: Looking for "Future Directions" in a Discussion Section https://youtu.be/2mTb6ydK3I4

Extra Video: Summarizing the Findings of an Empirical Study https://youtu.be/9UbLjwJRVNM

On the quizzes, you'll need to find opportunities for future research that the author(s) discusses. Remember that these are ideas that the author includes, sometimes tied to limitations, for what could be done next or differently to further expand our understanding of this area.

This is an example of translational research. This is a requirement for this journal, but it is a cool thing that they do. They put it under the abstract on the main webpage so that anyone can read it, even without full access to the article.

Extra video on contextualizing findings: https://youtu.be/UT61MWs5M4w

Extra video on Cronbach's alpha: https://www.youtube.com/watch?v=5gA2-MtWr8I

Who was studied: U.S. tweens between the ages of 8 and 13. Eligibility criteria were that participants had to be U.S. parents of a child in this age range — but notice the study is actually about the kids, not the parents. Parents were recruited as the gateway because the kids couldn't legally consent on their own. Out of 298 parents who initially clicked the survey link, 251 parent-child dyads completed the study (3 children declined assent, 13 children quit early, and 31 parents didn't complete the consent process). Notice the dropout funnel — that screening matters for the eventual sample. Let's break down the three layers:

Population of interest: U.S. tweens ages 8–13 (especially those who watch online video content like unboxing videos). Sampling frame: Active members of Dynata's U.S. online research panel who happen to be parents of tweens. Dynata is a research panel where people sign up to take surveys in exchange for points convertible to cash, travel miles, or prizes. Sample: The 251 parent-child dyads who actually completed the study.

Each step narrows the pool and potentially introduces bias. The population is all U.S. tweens; the sampling frame is parents who happen to be Dynata panelists (people who voluntarily sign up to take paid surveys); the sample is the subset of those whose kids agreed and completed the study. The further the sampling frame is from the actual population, the more bias creeps in — and Dynata panelists skew in particular demographic ways (whiter, more educated, higher-income than U.S. parents overall, as Table 1 shows). What's the sampling method? The authors don't explicitly name it, but this is a convenience sample (a type of non-probability sampling). They recruited from an existing pool of available panelists — not a random sample of all U.S. parents. You could also argue it has elements of purposive sampling because they specifically targeted parents of tweens (a defined characteristic), but the underlying recruitment is fundamentally convenience-based. Advantages of this approach: fast, inexpensive, access to a large pool, can screen for specific characteristics (like having a child in the right age range), can target by geography (U.S.-only). Disadvantages: not representative of all U.S. parents/tweens, Dynata demographics skew in particular ways (which the authors acknowledge in their limitations), self-selection bias (people who join paid research panels aren't random), and people who use Dynata regularly may not be typical parents. This matters for generalizability — findings may not apply to all U.S. tweens, especially those whose parents wouldn't sign up for an online survey panel. Extra video on figuring out the sampling method: https://www.youtube.com/watch?v=rstREj9jZdg

This is reporting the internal consistency reliability of the perceived informative intent scale. Cronbach's alpha (α) is what researchers use to check whether a multi-item scale is reliable — basically, do the items "hang together" the way they should if they're all measuring the same underlying construct? Alpha ranges from 0 to 1. Conventional thresholds:

α ≥ .70 = acceptable α ≥ .80 = good α ≥ .90 = excellent

An α of .82 here means the three perceived informative intent items are reliably measuring the same construct. If alpha had been low (.40, say), it would mean the items aren't really tapping a single thing — they'd need to be split into separate measures or revised. You'll also see Cronbach's alpha for the perceived selling intent scale, except wait — only TWO items measure selling intent, and the authors report a correlation (r = 0.53) instead of alpha. Why? Cronbach's alpha is technically defined for 3+ items. With only 2 items, a correlation tells you the same information more cleanly.

Is 251 enough for this study? Let's think it through. In a factorial design like this one, what matters isn't just the total N — it's the N per cell. With a 3 × 2 design and 251 participants, you're roughly looking at 251 / 6 ≈ 42 tweens per condition. (The actual breakdown is a little uneven, but that's the ballpark.) A common rule of thumb for between-subjects experiments is at least 30 participants per cell to get reasonably stable estimates, and more is better. So 42 per cell is decent — not generous, but workable. The authors were able to detect statistically significant effects in some of their hypotheses, which suggests they had enough power for the strongest relationships. But here's where it gets interesting: many of the hypotheses in this study were not supported (H1, H4, H6b were unsupported; H3, H5 were partially supported). When hypotheses fail, you have to ask: was there really no effect, or did the study just lack the statistical power to detect a small effect? With ~42 per cell, this study could detect medium-to-large effects but might miss small ones. That's a limitation worth thinking about when you read the discussion. Larger samples give you more power to detect smaller effects — but they can also make trivially small effects reach statistical significance. Sample size always involves a trade-off, and there's no single "right" number. It depends on your design, the effect size you expect, and what you're willing to risk missing. Extra video on sample size in experiments: https://www.youtube.com/watch?v=v-dyn6tO5dQ

Extra video on figuring out the sampling method: https://www.youtube.com/watch?v=rstREj9jZdg

What research paradigm does this study align with? This is a post-positivist study.

Evidence: - Hypothesis testing. The study advances 6 specific, falsifiable hypotheses (H1–H6, with sub-hypotheses) plus 2 research questions, all stated upfront before data collection. Findings are evaluated against these predictions, with hypotheses described as "supported," "partially supported," or "unsupported" — that's the language of post-positivism. - Quantification. All variables are measured numerically — Likert scales for perceived intent (1–5), correct/incorrect for recall items (0/1), summed scores for total recall (0–2), and a single rating for purchase intention (1–5). Statistical tests (path analysis, Wald tests, p-values) determine whether relationships exist. - Experimental control and random assignment. The authors manipulated two independent variables and used Qualtrics to randomly assign participants to conditions — both classic post-positivist tools for isolating cause-and-effect relationships. - Theoretical frameworks that predict relationships. The authors draw on schema theory and the Persuasion Knowledge Model to GENERATE testable predictions. Theory is used deductively (theory → hypothesis → test), not built up inductively from observations. - Search for generalizable patterns. The authors are interested in how tweens in general respond to unboxing videos — not how specific individuals interpret them. - Findings are reported as group-level statistics (means, regression coefficients), and the limitations focus on whether the sample is representative enough to generalize from. - Researcher distance from participants. Data collection happened through an online survey panel with no direct researcher-participant interaction. The setup deliberately minimizes the researcher's influence on responses — another post-positivist signature. - Acknowledgment of uncertainty. Note that this is POST-positivist, not strict positivism. The authors hedge appropriately — they discuss limitations, acknowledge that the manipulation might have worked through different mechanisms than predicted, and note ceiling effects and social desirability bias. They're seeking objective truth but acknowledge they can only approach it imperfectly.

On the quizzes, you'll be asked to find when the author(s) discuss ethical considerations explicitly and implicitly. For implicit, for example, authors may not mention some ethical considerations because it is obvious to other researchers that common ethical considerations were taken. In each annotation, use the hashtag #ethicalconsideration and explain what the author(s) did and why, using course concepts related to research ethics. These will be in the methods section and possibly in the discussion/conclusion section. AND because there are limited ethical discussions in this particular study, please reply to this annotation with some suggestions about what could have or should have been done, ethically.

Extra video on finding IVs and DVs: https://youtu.be/SlljkpUY4J4

Extra video on finding the bigger goal in a study https://www.youtube.com/watch?v=9e-pcABSu2Q

A Wald test asks whether two estimates differ reliably from each other. The authors use it here to test moderation: is the effect of perceived intent on the outcome significantly different in the training condition vs. the no-training condition? If the Wald test is p < .05, the moderation is real. If it isn't, the apparent gap between conditions could just be noise. You don't need to compute it - just recognize that when authors report a Wald test in a moderation analysis, they're testing whether two coefficients differ, not whether either one is significant on its own.

Article: Vaala, S. E., Mauceri, F., & Connelly, O. (2024). U.S. tweens' reactions to unboxing videos: Effects of sponsorship disclosure and advertising training. Journal of Children and Media, 18(2), 272–292. https://doi.org/10.1080/17482798.2024.2338541

Purpose: This annotated example is designed to help you prepare for the in-class experiment study identification quiz. Read through the article and these annotations carefully. The annotations explain what to notice and why — the quiz will ask you to do similar analysis on a different experiment study.

How to use this: Read the article section by section. When you encounter a highlighted passage, read the annotation. The annotations teach you what to notice and why it matters. This is NOT a graded assignment — it's a learning tool. When you take the in-class quiz on a different experiment study, you'll need to: - Identify the bigger research question and why the study matters - Identify hypotheses/RQs with their IVs, DVs, mediators, and moderators - Identify the sampling method and its strengths/weaknesses - Identify how participants were assigned to conditions, and the strengths/weaknesses of that approach (this is separate from sampling!) - Identify the type of experiment — true, quasi, or pre-experiment; lab, field, or natural; posttest-only, pretest-posttest, or time series; between-subjects or within-subjects; factorial or single-factor - Find and annotate ethical considerations, including those specific to research with minors (parental consent, child assent, IRB scrutiny of special populations) - Analyze variables: conceptual definition, operationalization, level of measurement, role in hypotheses, reliability, and whether each was manipulated or measured - Identify the research paradigm with supporting evidence Find limitations and future research directions Find places where authors connect their findings to previous research - Evaluate author and journal credibility - Apply experiment-specific concepts: manipulation checks, attention checks, cover stories, deception, demand characteristics, threats to internal validity, and the trade-off between internal and external validity - For each hypothesis, evaluate how the authors attempted to establish causality (association, temporal order, non-spuriousness) and whether they succeeded — paying particular attention to whether the IV was manipulated or measured

Remember to use your strategic reading skills: skim the abstract and Impact Summary first, map the structure using headings, then read the Methods and Results carefully. The introduction/theory can be skimmed for main ideas. The Discussion needs careful reading for limitations, future research, and connections to previous work.

Practice your strategic reading here. Before reading this article front-to-back, try this: Read the abstract (1 min). Skim the headings to see the structure (30 sec). Jump to the Methods - figure out: who was studied, how, and what was measured (5 min). Then read the Discussion for findings and limitations (5 min). THEN go back and read the Introduction/Lit Review for theory. This is more efficient than reading linearly, and it's the approach you should use on the quiz.

A Wald test asks whether two estimates differ reliably from each other. The authors use it here to test moderation: is the effect of perceived intent on the outcome significantly different in the training condition vs. the no-training condition? If the Wald test is p < .05, the moderation is real. If it isn't, the apparent gap between conditions could just be noise. You don't need to compute it - just recognize that when authors report a Wald test in a moderation analysis, they're testing whether two coefficients differ, not whether either one is significant on its own.

Here's a real example of what the peer review process looks like in practice. Let's trace the timeline for this article:

March 27, 2023: The author submitted the manuscript to the journal. Somewhere between March 2023 and March 2024: The editor sent it to peer reviewers (likely 2–3 experts in children's media, advertising literacy, or developmental psychology). Those reviewers read the full study and wrote detailed feedback — pointing out weaknesses, suggesting additional analyses, questioning interpretations, recommending additional literature, etc. The authors then revised the manuscript in response to that feedback. March 27, 2024: The revised manuscript was resubmitted (exactly one year after the original submission, which is on the longer end of the spectrum — it suggests substantial revisions). March 31, 2024: Accepted, just four days after the revision. That fast turnaround tells us the editor was satisfied that the revisions addressed reviewers' concerns and didn't need to send it back for another round.

So the total process from submission to acceptance was about a year. This is fairly typical for communication journals — some are faster, some take over a year. The key point: this study went through expert scrutiny before it was published. That's what distinguishes peer-reviewed research from a blog post, a news article, or a preprint. On the quiz, you may or may not have these dates for every study, but when you DO see them, they tell you something about the rigor of the review process.

A Wald test asks whether two estimates differ reliably from each other. The authors use it here to test moderation: is the effect of perceived intent on the outcome significantly different in the training condition vs. the no-training condition? If the Wald test is p < .05, the moderation is real. If it isn't, the apparent gap between conditions could just be noise. You don't need to compute it - just recognize that when authors report a Wald test in a moderation analysis, they're testing whether two coefficients differ, not whether either one is significant on its own.

Effect size is a measure of the strength of the relationship between variables in a study. It helps to understand the practical significance of the research findings.

A sampling frame is a list or database from which a sample is drawn. It should accurately represent the population being studied to ensure valid results.

Operationalizations refer to how abstract concepts are defined and measured in a study. It details the specific procedures or instruments used to measure variables.

The H index measures the productivity and citation impact of the publications of a researcher or journal. A higher H index indicates a higher number of influential articles.

A conflict of interest occurs when a researcher's personal, financial, or professional interests could influence their research outcomes. It's important to disclose these to maintain credibility and trustworthiness.

Peer review is a process where experts in the field evaluate a research paper before it is published. It ensures the research is credible, accurate, and of high quality.

A meta-analysis is a type of systematic review that combines the statistical results from multiple studies to draw broader conclusions. It provides a more powerful analysis by aggregating data from many sources.

Systematic reviews summarize the results of multiple empirical studies on a specific topic. They help to identify overall patterns and provide a comprehensive understanding of the research area.

Empirical studies are research studies that collect and analyze data to answer specific questions or test hypotheses. They are based on observed or measured phenomena.

A literature review is a comprehensive survey of existing research and studies on a particular topic. It helps to understand what has already been discovered and identifies gaps or areas for further research.

Intersectionality is the concept that various forms of social stratification, such as race, gender, and class, do not exist separately but are interwoven.

Essentialism is the belief that certain categories (e.g., gender, race) have an inherent, unchangeable essence.

Cronbach’s alpha is a statistic used to assess the internal consistency of a set of items, indicating how well they measure the same concept.

Intercoder reliability measures the extent to which different coders or observers produce consistent results when analyzing the same data.

The split-half method involves dividing a set of measures into two halves and comparing the results to assess reliability.

The test-retest method involves measuring the same individuals at two different points in time to assess the consistency of their responses.

Reliability refers to the consistency of a measure; a reliable measure produces stable and consistent results over time.

Content validity evaluates whether a measure covers the entire range of meanings within a concept.

Criterion-related validity assesses whether a measure is related to an outcome it should theoretically be related to.

Face validity refers to whether a measure appears to logically reflect the concept it is intended to measure.

Values are the numerical properties assigned to variables for statistical analysis.

Attributes are the different qualities or characteristics that make up a variable.

Variables are concepts that can take on different values in a study, as opposed to constants which do not vary.

Dimensions are aspects of a concept that group together logically, like subcategories within the overall concept.

Indicators are specific observations that show the presence or absence of a concept.

Operational definitions specify the exact procedures or measures used to observe or quantify a concept.

A conceptual definition is a theoretical explanation of what a concept means, similar to a dictionary definition.

Conceptualization involves defining the key terms and concepts used in a study to ensure clarity and shared understanding.

Code switching is the practice of alternating between different languages or dialects depending on the social context.

Otherness is a concept that describes how dominant groups marginalize non-dominant groups, making them feel excluded or diminished.

Operationalization refers to the process of defining and measuring concepts in a way that they can be empirically observed and tested.

Deficit-based thinking focuses on what marginalized groups lack or are deficient in, often blaming them for their circumstances. Asset-based thinking, on the other hand, focuses on their strengths and resources.

Cultural humility involves recognizing and respecting the cultural differences and perspectives of research participants. It means being open to learning from them and acknowledging your own limitations.

Process consent is the ongoing practice of checking in with participants throughout the research to ensure they still consent to participate and are not experiencing harm or stress.

These are the three core ethical principles from the Belmont Report. Respect means treating participants as autonomous agents, justice involves fairness in distribution of research benefits and burdens, and beneficence means minimizing harm and maximizing benefits.

The Belmont Report is a key document in the history of research ethics. It outlines ethical principles and guidelines for research involving human subjects, focusing on respect, justice, and beneficence.

IRBs are committees that review research proposals to ensure that they are ethical and that the rights and welfare of participants are protected. They are crucial for maintaining ethical standards in research.

The Hippocratic Oath is an ethical code attributed to the ancient Greek physician Hippocrates. It is still used today by medical professionals to swear to practice medicine ethically and honestly.

Satyagraha is a form of nonviolent resistance or civil disobedience. It was developed by Mahatma Gandhi as a way to fight injustice without using violence.

Ahimsa and satya are important ethical principles from Eastern philosophies. Ahimsa means non-violence, and satya means truth. These principles guide ethical behavior and decision-making.

Reflexivity is the practice of reflecting on your own values, biases, and influence on the research process. It's about being aware of how your background and perspective might shape your research.

The principle of objectivity in research aims to minimize the influence of the researcher's personal biases and ensure that findings are based on observable and replicable data.

Aggregation of numerical data points means combining individual data points to form a summary, such as an average or frequency, to identify patterns in the data.

Deductive reasoning involves starting with a general theory or hypothesis and then collecting data to test it. It contrasts with inductive reasoning, which starts with specific observations to develop broader generalizations.

The critical paradigm focuses on examining and challenging power structures and inequalities in society. It aims to use research as a tool for social change and justice.

Constructivism is a research paradigm that believes in multiple realities shaped by individual experiences and interpretations. Researchers are seen as active participants in the research process.

Positivism is a research paradigm that assumes a single, objective reality that can be discovered through scientific methods. It often involves generating and testing hypotheses.

Community-based research involves collaborating with community members to address local issues. It can use quantitative, qualitative, or mixed methods.

Mixed methods research combines both quantitative and qualitative approaches in a single study to provide a more comprehensive understanding of a research problem.