Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Sample size problem while using logit


I am conducting a small scale research using CBC. There are 20 respondents and each were ask to answer 15 choice tasks (4 attributes, 5 levels each + none dual response). My question is whether it will be possible to use logic estimation? I am not sure how the software works. I am thinking about running a regression on other program and how many observation I really get, 20 (as the number of respondents) or 300 (20*15 as the number of tasks done)?

This is my first time using this program. Please feel free to let me know if you have any suggestion.

asked May 15, 2014 by Komson

1 Answer

0 votes
Hello, Komson.

Your sample size will be 20 x 15 = 300.  Your model has 17 parameters to be estimated, so you will have about 15 observations per parameter.  This is thin for a logit model but the model should estimate for you.

However, basing a model (any model) on 20 respondents is risky:  this is not a function of logit but rather of any kind of research whatsoever (it would affect a regression model or correlations or any other statistical model).  Unless the universe you are trying to generalize to is extraordinarily small (e.g. 30) then your results will be subject to very large sampling error.
answered May 16, 2014 by Keith Chrzan Platinum Sawtooth Software, Inc. (53,875 points)
Hello Keith,

Thank you very much for answering my question. I understand about the sample size issue but there is nothing I can do.

Another question if you could answer is what should be the degree of freedom for the t-statistics shown in the result? Should it be based on 300 observation or 15 observations then? I am a newbie user of this program so I hope you don't mind these questions.