Undertaking modeling and inference and subsequently apologizing for analysis deficiencies results in scientific noise
Consider the edit: "Undertaking modeling and inference and subsequently apologizing for deficiencies in the data results in scientific noise"
Rather than accepting or not accepting a proposed model on the basis of a GOF assessment, embed the proposed model inside a more general model that relaxes the assumptions, and use AIC or a formal test to decide between the two. Comparing only two pre-specified models will result in minimal model uncertainty.
As a more general expression or recommendation, this sentence might be useful to insert somewhere: "It is often more useful to think of assessment of goodness of fit in terms of a contest between the proposed model and a more general model."
Perhpas the sentence could be inseted as the second bullet.
The potential magnitude of signal, penalized for possibility of accident. Or— the potential magnitude of signal, adjusted for risk of coincidence.
? Should this be simplified to: 'The potential magnitude of signal, penalized for possibility of accident or risk of coincidence.'
All standard regression models have assumptions
This is just a note elaborating the clause (not a suggested correction).
Fundamentally, a statistical model, per se, is a set of assumptions (constraints on possible conditions about the data generating process) that permit us to calculate estimates that we believe will properly represent or predict phenomenon of interest.
assessing and satisfying assumption types 1, 2, and 4.
Where these assumptions (linearity [1], additivity [2], proportionality [4]) are found in assessment not to obtain, methods for relaxing or modifying assumptions (e.g., non-parametric covariate expression, or interaction terms) to support accurate model estimation and prediction are presented.
Note that the only efficient full likelihood software implementation at present (besides SAS JMP) for continuous Y in large datasets is the orm function in the R rms package, which can efficiently handle more than 6000 intercepts (one less than the number of distinct Y values).
Is it the size of the data, per se, that is the impediment to implementing in standard SAS, or the large number of intercepts? Could ordinal regression be performed for large number of intercepts for a continuous Y on a data set of moderate size (e.g. 300 to 500)? Does repeated measures in longitudinal ordinal models complicate running models in packages other than R, such as SAS?
females
I believe 'for "males"' intended here.