What do I look for in a survey dataset as a (crime survey) data user? These slides are from the Survey Futures Survey Practice Forum #4 held on 6 December 2024. The talk covered some of the strengths and challenges of using the Scottish Crime and Justice Survey to measure changes in victimization over time.
The Tyranny of the Means Regression coefficients focus on mean differences between groups. This is fine. But sometimes we want more…
What to do with an odds ratio of 35? With matched case-control data (e.g. matching each ‘case’ of your outcome to three ‘controls’), researchers are recommended to fit conditional logistic regression models to account for the structure of this matching in their analysis. However, this means that by default you can’t calculate meaningful marginal effects or predicted probabilities from your model - because the model’s intercept is determined by the ratio of cases to controls. This leaves you to focus on odds ratios… but what if you (or your audience) don’t like the odds ratios you get? Using an example from ongoing analysis from the Policing the Pandemic in Scotland project, in this presentation I will talk through our thought process in figuring out what we can do with some odds ratios we don’t really like, and end up asking what is a marginal effect anyway?