choice-metrics.com

Dear Sir,

I have prepared the choice scenario for the case having 5 attributes (with 3 levels and a few with 2 levels) and 5 alternatives using Ngene. I have adopted an efficient design technique to design. The Ngene results in 60 choice scenarios with divided into 10 blocks. One thing I didn't understand is whether we can randomly assign the 60 choice scenarios to respondents, giving each respondent 6 scenarios, without considering the block numbers. Is it mandatory to consider the block numbers and follow the same block numbers for every scenario?

Also,
I got the
D error 0.202252 (is this model is efficient, what will be the minimum d-error to be achieved)
A error 2.179224
B estimate 89.061645
S estimate 126058.409 - This sample is huge...

How to achieve the sample size less when we target for 1000 sample size...

Yes you can randomly assign choice tasks if you like, that is what most survey platforms do. The main benefit of blocking is that each block has reasonable attribute level balance such that each attribute level appears more or less equally in the choice tasks given to each respondent. With random assignment of choice tasks it is possible that one respondent only sees low levels of price, for example. But it usually does not have a major impact on model estimation.

The S-estimate is large because you probably provided a near-zero prior for one of your parameters. if it is near-zero then the attribute level has likely no effect and its effect will be difficult to measure unless you have a very large sample. You should look at S-estimates of individual parameters instead of the overall S-estimate.

Michiel

Thanks for the reply professor..

If we assign randomly (6 scenarios to each respondent), and we are targeting around 1000 sample size. Is there any possible minimum number of entries that should be there for each scenario? Since, it will be a online survey, we don't have control over the distribution of scenarios to respondents. For example, scenario number 5 (out of 60) may contain 10 entries and scenario number 20 may contain only 2 entries. Does this have any impact on our model estimates and results.
Do we have to maintain certain minimum number of entries for each response?

Thank you again for the reply.

In the unlikely event that one specific choice task appears much more in the data then another choice task then your estimates will be more based around some attribute levels than others, which means that forecasts away from these attribute levels are less reliable. However, it is unlikely to affect your parameter estimates much because all choice tasks should measure the same behaviour. I would not worry about random allocation of choice tasks too much. Usually there is no minimum number for each choice task to appear in the data, but if your number of rows in the design is very small I would recommend to get at least 1 observation per choice task to ensure that all parameters are identifiable. But if you are using a sufficiently large number of rows in your design then it does not matter.

Think about RP data, where each choice set is likely to be different from any choice set. You have no control over choice sets in RP data and you can still estimate models without a problem.

Michiel