Bootstrapping pairwise not listwise, a problem with N


Hoping someone more statistically minded can help.

I have a number of variables with different N (they are performance measures and some have been removed due to not understanding the task). I want to bootstrap the correlations but when I do this in SPSS it excludes listwise rather than making use of all the data available (i.e. pairwise). I am wondering if one workaround is to bootstrap each pair of correlations rather than producing the matrix of all variables? My decision on what to include in the regression model would be based on these correlations. So my next but related question is, is different N for the variables a problem for regression?

Many thanks!


Less is more. Stay pure. Stay poor.
Yes, most regressions will listwise delete an entire observation if they are missing a data point for any variable in the model. You may need to evaluate whether excluding these observations will change or bias your results. So you need to understand your missingness in order to decide if something like multiple imputation may be useful.