I'm analyzing an archival data set for my dissertation and I can not find in my statistics books a citation indicating the number of cases appropriate for each independent variable. I am running a multiple regression analysis with 6 independent variables (college engagement factors) and one dependent variable (time to completion)
Peter Samuels
Dear Karina Clennon ,
There is a discussion on page 313 of Field, A. (2013) Discovering Statistics Using IBM SPSS Statistics: And sex and drugs and rock ‘n roll. 4th edn. Which I summarise as:
- There are various rules of thumb circulating about the required ratio of the sample size (N) to the number of variables (k) but these are too simplistic, apart from the simple requirement N to be as large as possible.
- We also require N > k.
- The expected value of |R| for MLR with random data is k/(N-1). Therefore, if k is fairly large relative to N, there will be a high expected value of |R|, so it will be difficult to detect a real effect from an random effect.
- Field suggests the best way to approach this is as a Type II error sample size calculation. This requires an a priori estimation of the effect size, |R|, or R2.
He then provides a table of sample size calculations derived from G*Power.
0 votes
0 thanks
David L Morgan
There are a variety of online resources for sample size calculation. For your case, I would recommend the program G*Power.
2 votes
1 thanks
- Badges
- Science topic
- Similar topics
- Citations
More Karina Clennon's questions See All
Similar questions and discussions