The authors do not go into depth about this choice, but considering all respondents are receiving cash assistance, why do you think the authors may have chosen this measurement as a scale of economic status?

This finding makes sense in that having one's utilities shut off is probably not an indicator of being sanctioned from the TANF rolls.

The phrase "pseudo R-squared" is going to show up a lot in the next few pages. Because the dependent variable(s) in a logistic regression model is not continuous, R-squared cannot be computed in the same way using ordinary least squares and minimized variance cannot be assumed. That being said, psuedo R-squared does not imply that inferences cannot be drawn from these methods or that these methods are inherently unreliable.

This paragraph and the one that follows outline the sanctioning process in Michigan fairly clearly; sanction criteria and mandates for benefit recipients are well defined for the state. For this type of analysis, could this be a potential advantage?

Do you think this is a suitable sample for this study? Given the authors' emphasis on variability between states in their policy implementation, how do you think this will affect their analysis? Are their any other sources you can think of that might be more suitable if so?

Some more background: this policy, implemented under Clinton, was devised to reach across the political aisle. The idea was to implement a wide reaching social safety net with a "tough love" attitude that would appeal to the political right.

Some background; PRWORA changed how cash benefits were distributed by implementing a block grant system in which states were delegated grants and then allowed to allocate that money in ways that they saw as most beneficial to their electorate. As you can imagine by the tone of this article, this was implemented with various degrees of success.

Some background about why the changes made under PRWORA are of note. This in addition to the implementation of a block grant cash assistance program and the term limit of benefits are what makes the comparison between PRWORA and AFDC worth investigating.

I'm not convinced that a chi-square test can support this kind of broad conclusion. At the very least, the wording is a bit loaded in my opinion. All a chi-square test can really say in this case is that both of these variables have /some/ effect on their science funding variable, but teasing out the extent of that correlation cannot be determined from a Wald test alone.

Which likelihood ratio test? Are they talking about an LRT proper, a chi square test, and F-test?

Is this the author saying that they think that the distribution of this variable is normal? It's my understanding that ordinal logistic regression assumes some degree of variance between ordinal values that can be explained by other variables. If the author assumes a nonnormal distribution, how do they justify using z-score transformations for their independent variables?

An interesting approach! Typically I see these types of variables recoded and fit into a logistic model as dichotomous variables. Reading a bit about this approach, it seems like that is essentially what is being done, just with different regression coefficients for each group.

Without the original documentation, does this seem like an appropriate operationalization of group membership? Is it too broad/narrow?

Here we see that the authors used a purposive sampling method when recruiting participants. On the one hand, given that only 3% of the population identifies as "animal product limiters", this method helps ensure a decent sample size rather than random sampling. On the other hand, the fact that participants are already involved in circles formed around the "animal product limiter" identity, this may conflate some of the authors' assumptions, particularly relating to the influence of a vegan or vegetarian "group" on maintaining those lifestyles. Given the scope of this research, do you consider this an appropriate sampling method?

It looks like the authors omitted "oral health" from their battery of factors in this case.

FCQ = "Food Choice Questionnaire". The FCQ is a measure that assesses 10 main factors relating to peoples' food choices; health, mood, convenience, sensory appeal, natural content, price, weight control, familiarity, ethical concern and oral health.

Given this focus of the research study, what sort of statistics do you expect to see?

Here I believe "Tukey comparisons" refers to Tukey's HSD (honest significant difference) procedure. The main difference between a Tukey HSD and a t-test is that comparisons are drawn from the pooled variance rather than the variances of two groups. This statistic is designed to mitigate the risk of finding a "false positive", or type 1 error, that may arise when performing a multitude of pairwise functions.

Do the authors ever explain the significance of having left a job in the last three years? The only explanation I can think of is respondents' limited recall ability about workplace events that might skew the results. I wonder how this subgroup analysis would look had they included respondents who had left a job in the past five or ten years.

The authors have a strong conceptual framework for investigating victimization based on past research, on adults but lack a strong theoretical framework for why adolescents specifically fear victimization/have a higher degree of fear of being victimized. Hence the purpose of this study. They suspect that adults and adolescent fear of crime are characterized by the same factors, but the literature has lacked a representative sample of adolescents thus far.

Note that in these models, the nodes location does not denote their centrality measures (unlike some models we may have seen/used, such as Yifan Yu, Force Atlas, or Fruchterman-Reingold). The model is intentionally chosen to create distinct visualizations of subgroups.

Gephi's Yifan Yu model is an example of a force directed model, for reference

Identifying nodes that play a similar role in different subgroups

This process may be of interest for the large network project; extracting information from http protocols can be done relatively simply using Python's beautifulsoup library (https://en.wikipedia.org/wiki/Beautiful_Soup_(HTML_parser)).

Given the nature of the network being analyzed, I wonder how useful a lexical, name-based extraction would be at identifying useful information ,i.e. how likely is it that these bloggers are using their real identities or pseudonyms that would be properly extracted? What special considerations have to be made when algorithmically extracting network specific data?

An interesting perspective I hadn't considered before; weak ties speak more to the structure of the entire network than to the relationships between the people who share them. Paired with their method of quantifying relationship strength, this allows for a two-level analysis to address their multiple hypotheses.

Without being fully embedded in the network, actors can help circulate new, novel information or connections. Weak ties != bad ties, and in this case help less prestigious actors find clout within the network.

Especially if these organizations bonded together in opposition to organizations that vilified them prior to 1980, I wonder how this network is structured compared to other formal religious networks forged under different circumstances. Did this work ethic precede the formation of these political institutions, and therefore exacerbate them?

This seems to go back to the question of whether social networks reify social capital or help equal the playing field. In light of their observation that conversations occur most frequently between users of similar "status" and the high correlation between centrality measures and twitter follower count, are users viewing their interactions as collaborative in the way that the author's do? Or is it the case where the structure of physical networks is just being replicated online? As the author's note below, most tweets in this sample were used as a way of disseminating information, but without a strong core component, this could be the case where misinformation could be spread without any outside observers intervening.

It could be that this is less an observation about the network and more of an observation about the medium. Low reciprocity among connected users is pretty par for the course on Twitter (https://www.cs.bgu.ac.il/~snean151/wiki.files/22-WhatisTwitterASocialNetworkOrANewsMedia.pdf). Predictive models have even been constructed to determine how likely two users are to engage in reciprocal behavior on Twitter to address this concern (https://www.cs.cornell.edu/home/kleinber/socialcom11-recip.pdf)

It could also be the case that this finding is partially due to their omission of quoted tweets, in which reciprocity could have occurred. Or perhaps Twitter's Direct Messages are where most of the collaborative conversations occur, meaning we may have to go beyond the API to get such information.

In light of the findings of their logarithmic analysis, why not a multi-network analysis of different occupations to see differences in how that knowledge travels versus knowledge ascertained from mass media?

Maybe this could presuppose a longitudinal analysis of how information sharing changes based on the availability of mass media, ie how the structure of a network changes as mass media becomes more prevalent with IEK or vice versa. Or, if IEK can influence disaster mitigation or resource management efforts, what affect would these efforts have on IEK?

This makes it sound as though the edges between the players were all weighted the same. I wonder how these networks would look had the responses been listed for each player accompanied by a Liekurt scale for measure, rather than a binary measure. For example, if x feels confident in y's basketball ability but feels confident in z's ability to a lesser degree, how does that affect the network structure?

To what disciplines do these 'state subsidized entrepreneurs' belong? Does this model resemble a tree in which the private corporation is connected to the university which is then connected to the individual, or do these 'entrepeneurs' create a point of contact between these corporations and the university?

What kind of significant differences? This might suggest a network model of scholars, with edges drawn between scholars the same discipline. Edges could also be drawn between scholars who worked on the same research projects. Node size could signify the amount of funding received to see where this funding is concentrated the most.