Checking assumptions using complex samples

[Sarah Wilson]

Hello. I am working on a project running an ANCOVA using complex samples. I am afraid my data violate the normality and homogeneity of variance assumption. Are there any recommendations on how to check assumptions using complex samples?

[Carlo Lazzaro]

Sarah:
welcome to the list.
This kind of question are at risk of being left unreplied, due to a substantive lack of details,
Please, take a look at the FAQ on how to post (more) effectively. Thanks.

[Steve Samuels]

And, please, tell us about the study design. Was the sampling was planned with this analysis in mind, perhaps by over-sampling one of the comparison?

[Steve Samuels]

Non-normality and heterogeneity do not matter for survey analyses. Survey standard errors are based on between PSU variation and are robust to distributional assumptions. What is important is specifying a good model, e.g.assessing the need non-linear predictors or interactions.

I asked about the sampling design because there is disagreement among sampling experts about whether sampling weights are needed for causal analyses. If you have oversampled one subgroup to ensure equal numbers for an ANCOVA, then I would notweight. Otherwise, I suggest that you compare weighted and unweighted versions to see if it matters.