Using Constants? Is it possible with Sawtooth?

Question

Hi,

Often in models, particularly nested logit models which have large gaps in their experimental designs, or models that have highly restricted designs (such as transport mode choice) it is common to use a constant to stabilize an estimation as the constant accounts for all the unexplained heterogeneity in a particular (usually alternative specific) utility estimation. Sawtooth does (obviously by design) have an option to include a constant term in the utility estimation procedure, why is this the case? and is there a "power" trick that can allow you to include a constant. Many applications/journals/fields of study expect this as standard. Can I adapt my sawtooth estimations to help account for this?

Many thanks

J

Answer 1

Jasha,

Yes, you can include alternative-specific constants (ASCs).  In our software it will often be estimated as the utility of the attribute upon which the alternative-specific effects depend (e.g. brand or transportation mode).  

On occasion, for particular design situations (e.g. when I have more than one constant fixed alternative), we need to add ASCs manually.  The power trick here is to export your design from our software as a .csv file,  Then work in Excel to add columns to the design to capture the ASCs of interest.  You can then import your customized design into our stand alone CBC/HB software, use menus within the software to make sure your ASCs are properly coded (typically by identifying them as "user-specified coding") and then you can run your models.