Have an idea?

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

Is the standard error in conditional logit output actually standard deviation?


I would like to ask if the standard error in the CBC logit output is actually standard deviation and if we need to divide it by square root of the sample size, in order to obtain the standard error?

The background to this question is we want to find out the marginal willingness-to-pay (WTP), using this equation:

mWTP = Difference in utility between two levels/gradient of the graph utility against cost

To find the 95% confidence interval of mWTP, we need the standard error of the difference in utility between two levels.

Thanks for your help!
asked Dec 22, 2016 by Germaine

1 Answer

0 votes
The standard errors are standard errors of the estimation.  If you increase sample size by a factor of four they will decrease by half during estimation, so there is no need to make an additional division by the square root of sample size.
answered Dec 22, 2016 by Keith Chrzan Platinum Sawtooth Software, Inc. (50,900 points)
Thanks for the answer!

In this case, if I want to find out the standard error for the difference in utility between two attribute levels, may I know what method or formula should I use?

You could use the same formula you would use for any other stat test, like a t-test.  Differences in the part worths between the two levels in the numerator, the square root of the sum of the two squared standard errors, right?  Note that it is not appropriate to compare levels from different attributes this way, and the test is a weak one because it is an independent groups T-test, not a dependent t-test, which would be more accurate.   If you sent me an email at keith@sawtoothsoftware.com I can share additional information if you want to run this test with HB utilities instead of aggregate MNL utilities.
Ok, thank you so much. Will send you an email shortly.