levels range- quick quest.

[Anat Tchetchik]

Dear all,

To unveil people's pref. toward the use of their old cell phones, we created the following labelled design which is based on real prices in new Israeli shekels (in $1=NIS4).
I wonder if the price range is not too big. Also does the use of 4 levels should provide better statistical results (significant), otherwise I would use 3 levels and do a full fact without blocking.

Design
;alts = Resyclcomp, yad2, tradeoff, donate, giveaway, throw, leavehome
;rows = all
;fact
;block=8
;model:
U(Resyclcomp)= a1 +b2*priceA[145,245,345,445] /
U(yad2) = b1+ b2 *priceB[542,700,990, 1203] /
U(tradeoff)=c1 +b2*pricec[155,255,355,455]/
U(donate)= d1 /
U (giveaway)= e1/
U(throw)= f1$

Anat

[Michiel Bliemer]

Dear Anat,

I cannot tell you if the range is too large; if the choice tasks make sense to you, then it is probably OK. If you feel that price is too dominant in relation to the other attributes, then maybe the range is too large.

Using 3 levels is usually statistically more efficient than 4 levels. The most efficient design (given that the range is not too large) is an end-point design in which only 2 lev els are used for each attribute. But clearly it will not be possible to test for nonlinear effects in estimation.

Michiel

[Anat Tchetchik]

Thank you very much Michiel.
Just to be sure, with 3 levels it should be possible to test for nonlinear effects in the estimation?
Anat

[Michiel Bliemer]

With 3 levels you can estimate 2 dummy or effects coded parameters. So yes you can test for nonlinearities, but only 2 coefficients. With 4 levels, you can estimate 3 dummy or effects coded parameters, so you can test more elaborately for nonlinearities.

[Anat Tchetchik]

OK, Thanks much!