Dear all,
I am running a nested logit model on an unlabelled discrete choice experiment. My nests are: policies (3 options) or no policy (1 option). If I run the nested logit model only with the attributes, the results are as expected (people are more inclined to choose policies that cost less, lead to a higher share of recycled materials, a higher reduction in greenhouse gas emissions, and a higher share of reusing - the context are policies to increase sustainable building in cities). Now when I add a variable that measures how important people think sustainability is in general (N001_01), the signs change to the complete opposite, which I cannot explain. Why would people choose the policies that cost them more? It seems to me that this is an artifact, but I have no explanation.
Here is the output (first without, then with inclusion of the sustainability variable):
I am running a nested logit model on an unlabelled discrete choice experiment. My nests are: policies (3 options) or no policy (1 option). If I run the nested logit model only with the attributes, the results are as expected (people are more inclined to choose policies that cost less, lead to a higher share of recycled materials, a higher reduction in greenhouse gas emissions, and a higher share of reusing - the context are policies to increase sustainable building in cities). Now when I add a variable that measures how important people think sustainability is in general (N001_01), the signs change to the complete opposite, which I cannot explain. Why would people choose the policies that cost them more? It seems to me that this is an artifact, but I have no explanation.
Here is the output (first without, then with inclusion of the sustainability variable):
HTML Code:
nlogit Choice Costs RecMat GHG Reuse || nests: , base(none) ||Alternative:, noconstant case(id)
...
RUM-consistent nested logit regression Number of obs = 7,536
Case variable: id Number of cases = 1884
Alternative variable: Alternative Alts per case: min = 4
avg = 4.0
max = 4
Wald chi2(4) = 768.86
Log likelihood = -1685.8381 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
Choice | Coefficient Std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
Alternative |
Costs | -.0645158 .004922 -13.11 0.000 -.0741628 -.0548687
RecMat | .0187533 .0014563 12.88 0.000 .0158989 .0216076
GHG | .0205733 .0009328 22.06 0.000 .0187451 .0224015
Reuse | .0125099 .0011097 11.27 0.000 .0103349 .0146849
------------------------------------------------------------------------------
dissimilarity parameters
------------------------------------------------------------------------------
/nests |
policies_tau | .5548908 .0298003 .4964834 .6132982
none_tau | 1 . . .
------------------------------------------------------------------------------
LR test for IIA (tau=1): chi2(1) = 107.72 Prob > chi2 = 0.0000
HTML Code:
nlogit Choice Costs RecMat GHG Reuse || nests: N001_01, base(none) ||Alternative:, noconstant case(id) constraints(1)
...
RUM-consistent nested logit regression Number of obs = 7,536
Case variable: id Number of cases = 1884
Alternative variable: Alternative Alts per case: min = 4
avg = 4.0
max = 4
Wald chi2(5) = 567.41
Log likelihood = -1573.4881 Prob > chi2 = 0.0000
( 1) [/nests]none_tau = 1
------------------------------------------------------------------------------
Choice | Coefficient Std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
Alternative |
Costs | .017561 .0055667 3.15 0.002 .0066504 .0284716
RecMat | -.0051151 .0016289 -3.14 0.002 -.0083078 -.0019225
GHG | -.0055541 .0017301 -3.21 0.001 -.0089451 -.0021632
Reuse | -.0033491 .0010736 -3.12 0.002 -.0054534 -.0012449
------------------------------------------------------------------------------
nests equations
------------------------------------------------------------------------------
policies |
N001_01 | .0531828 .003809 13.96 0.000 .0457173 .0606482
-------------+----------------------------------------------------------------
none |
N001_01 | 0 (base)
------------------------------------------------------------------------------
dissimilarity parameters
------------------------------------------------------------------------------
/nests |
policies_tau | -.1497973 .0468949 -.2417096 -.057885
none_tau | 1 (constrained)
------------------------------------------------------------------------------
LR test for IIA (tau=1): chi2(1) = 332.37 Prob > chi2 = 0.0000
