Small sample of dichotomic variables

[Elias Carvalho]

I have a small sample (only 17 records) of 5 dichotomic variables and low variability for same variables like: 0 = 2, 1 = 15.

Questions:

1. Is it possible to create a path analysis model with this small sample? 

2. How to deal with this dichotomic data ? 

Some example please?

Thank you

[Terrence Jorgensen]

1. Is it possible to create a path analysis model with this small sample? 

Anything is possible, but with such a small sample and some variables having only 2 observations in category 0, your model will probably not converge if that variable has a predictor that allows perfect prediction of that outcome (look up the separation problem in logistic regression on Google or Stack Exchange).  Regardless, you will have very low power and high sampling variability, so any significant effects (true effects or Type I errors) will likely grossly overestimate the true effect size.

2. How to deal with this dichotomic data ? 

You can declare them as orderded with the ordered= argument, which will basically run a probit regression, but with intercepts fixed to zero and the thresholds estimated (instead of the usual case, as in the glm() function, which is to fixed the thresholds to zero and estimate the intercepts -- they are inversely equivalent).  Or just run separate probit or logistic regressions for each outcome using the glm() function, since a path analysis might not converge with only N = 17.  

Terrence D. Jorgensen

Postdoctoral Researcher, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

UvA web page: http://www.uva.nl/profile/t.d.jorgensen