Have an idea?

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

S.E's in Alt-Specific Designs

This is likely to be a very rudimentary question but I just want confirmation from the community/Sawtooth Team that when testing (advanced) Alternative Specific Designs that for an acceptably efficient design, the standard errors of the part-worth estimates (given the number of prohibitions used in the Alt-Spec Design) for the attribute levels of the named Alternative Alternatives (i.e. Car, Bus, Train) should be below 0.1 (with increasingly lower values being better) that is for Car, the attribute 'daily parking rate' which may have levels ranging from $10-$40, each incremental value should have a standard error of less than 0.1? I believe that is what the warning/call-out in the test result suggests but just wanted to confirm.

However, would it still be the case that the named alternative (i.e. Car, Bus, Train) estimate, we would still want to see SE's of 0.05 or lower. Would that be a correct assumption to make? Being they are first making a primary choice between these named alternatives as they are higher order attributes. Or is this an incorrect assumption.
asked Mar 6, 2017 by Jasha Bowe Bronze (1,745 points)
retagged Mar 6, 2017 by Walter Williams

1 Answer

0 votes
Best answer
Yes, in my experience, the primary attribute (the attribute containing the levels{ car, bus, train, etc.--also known as the ASCs) should have target standard errors of around 0.05 or less for simulated data.  But, the alternative-specific attributes (e.g. the prices for bus, or the frequency of trains) should have target standard errors of 0.10 or less.
answered Mar 6, 2017 by Bryan Orme Platinum Sawtooth Software, Inc. (154,405 points)
selected Mar 6, 2017 by Jasha Bowe