Have an idea?

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

ACBC covariate interpretation- alpha file

Thank you very much once more !
I am sorry that I keep asking, maybe very basic questions about it, but I couldn't understand this.

I could finally open the alpha file which gives me over 40000 iteraction. Is this a correct outcome? and I would be glad if you could help me with the interpretation of it regarding the affect of covariates.

I just want to see the affect of my covariates on the attributes and their levels. I have added 5 attributes to my survey in ACBC (without BYO, only choice tasks) and I want to know how the 3 covariates (which I used likert scales with 3,4, and 6 items)  effect their choices of attributes and levels and to what degree. I want to know how a person being individualist affects her choices and to what degree compared to collectivist person. Could you help me how can I do that?

Should I calculate the means of the items to come up with just 1 column to add on my data on Sawtooth HB estimation or should I add the items all as covariates ?

Should I use SPSS to find out the moderating roles ? or am I able to see that within the HB estimation output ?

I would be more than glad if you could figure this out as I am stucked now between softwares to find a way to see this.

Thank you again for your help and time.
related to an answer for: ACBC covariate estimation - alpha file
asked Dec 11, 2014 by Mir (400 points)
edited Dec 11, 2014 by Mir

1 Answer

0 votes
You will get a row in the .CSV per iteration of HB.  You'll want to decide which iterations to "throw away" because convergence wasn't achieved.  Typically, researchers throw away the first 5000 or first 10000.  Then, you only pay attention to the remaining iterations.  

You'll find the columns labeled in the .csv file, so you can see which columns refer to what.  You'll first have the alpha estimates (intercept) when the covariates are zero, and you'll have the columns referring to the adjustments to the alpha values according to the coded covariates.

It's easier to interpret the intercept alphas if you have specified continuous covariates that are zero-centered.

A typical procedure is to count for how many of the rows a particular covariate beta is consistently either >0 or <0.  For example, if your covariate were income, you will see for a particular part-worth utility the beta (either positive or negative) for each iteration.  If across 98% of the used iterations (after convergence is assumed) the income covariate is associated with a >0 beta, then you are 98% confident that income has a positive effect upon preference for that part-worth utility (for the sample).
answered Dec 11, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (142,615 points)
is this still applicable for covariates that are included as  likert scales ?
...