Analysis after dcreate

[Laurence Lester]

Dear List

I have used dcreate (thanks to Arne Risa Hole) to set the questions for my DCE. I have 9 choice sets; each as 2 options; each option has 3 unordered attributes (e.g. blue, green, yellow). I have collected the data, but am at a loss as to how to analyze the data.

I have checked all the examples of DCEs I can find but they all seem to be based on a single set (e.g. the carchoice example with 'help cmclogit,) or a choice between 2 options (with alternatives -- often ordered -- in them).

If it helps this is an example of one (of my 9) choice sets:

Choice set 1	Option 1	Option 2
Energy affordability	Solar Panels	All major appliances serviced
Quality and condition	New paint and carpet	Deep clean
Energy efficiency	Add or upgrade ceiling insulation	Thermal block-out blinds or outdoor shading

I would appreciate it if someone could guide me to an example of analysis of data that contains a set of questions that arise out of dcreate.

Thanks
Laurence

[Laurence Lester]

Of if somebody would kindly suggest which Stata model I should be looking at I could take it from there...Thanks. Laurence

[Hong Il Yoo]

A good starting point is McFadden’s conditional logit model (-clogit- and -asclogit- in Stata) and its modern extensions that account for unobserved heterogeneity (e.g., community contributed packages like -mixlogit-, -mixlogitwtp- and -lclogit2-).

[Laurence Lester]

Thank you Hong. That confirms what I though after much searching/reading (to put it in context; thanks as I have taken over a project that is late, has run out of funding, and the colleague who was to do the analysis is no longer able to be involved). Regards, Laurence

[ABD HILO]

Hong Il Yoo I am analyzing a discrete choice experiment data that I collected. When I coded the salary as a percent increase and as money value it gives me a different coefficient for salary and when I calculate WTP; for salary coded in percent increase it is in hundreds and those coded in money values it is in tens of thousands. here are the first eight rows of the data. I used Hole's (2007) mixlogit command. I calculated the WTP in preference space by allowing full correlation of the coefficients.

resond.ID	obsid	alt	salary(percent increase)	salary(money value)	Education	Location	Housing	Timely Payment	Workplace management and	Equipment and drug supply	Infrastructure	choice
15	1	1	0	4609	4	2	3	1	2	2	1	1
15	1	2	60	7374	3	1	1	2	1	1	2	0
15	2	1	80	8296	2	3	3	3	2	2	1	0
15	2	2	40	6453	1	2	1	2	2	1	2	1
15	3	1	60	7374	1	1	2	1	1	2	1	1
15	3	2	60	7374	4	1	2	1	1	1	2	0
15	4	1	80	8296	2	1	2	2	2	2	2	0
15	4	2	40	6453	3	2	3	3	1	1	1	1

//***List of attributes excluding constant***//
. global randvars1 "educ_3 educ_2 educ_1 housing_basic housing_adv district_center zone_center delay_amnth pay_ontime mgmt_support equipdrugs infrastructure cons"

. mixlogit choice salary, id(respond_id) group(pair) rand($randvars1) nrep(500) corr Where salary is coded as percent increase

Iteration 0: log likelihood = -3179.2124 (not concave)
Iteration 40: log likelihood = -3029.9196

Mixed logit model Number of obs = 10,048
LR chi2(91) = 335.18
Log likelihood = -3029.9196 Prob > chi2 = 0.0000

---------------------------------------------------------------------------------
choice | Coefficient Std. err. z P>|z| [95% conf. interval]
----------------+----------------------------------------------------------------
salary | .0093119 .0009229 10.09 0.000 .0075031 .0111208
educ_3 | .7575421 .0911441 8.31 0.000 .5789029 .9361813
educ_2 | 1.154959 .1006774 11.47 0.000 .9576354 1.352284
educ_1 | 1.413904 .1213084 11.66 0.000 1.176144 1.651664
housing_basic | .5007354 .0798207 6.27 0.000 .3442898 .657181
housing_adv | .6274487 .0798592 7.86 0.000 .4709276 .7839698
district_center | .1137681 .0681816 1.67 0.095 -.0198653 .2474016
zone_center | .2889555 .0733648 3.94 0.000 .1451632 .4327478
delay_amnth | .0233192 .0671155 0.35 0.728 -.1082247 .1548631
pay_ontime | .2132795 .0710018 3.00 0.003 .0741184 .3524406
mgmt_support | .3494241 .0563959 6.20 0.000 .2388901 .459958
equipdrugs | .1214757 .048055 2.53 0.011 .0272895 .2156618
infrastructure | .1700855 .0488378 3.48 0.000 .0743652 .2658057
cons | -.072252 .0615397 -1.17 0.240 -.1928677 .0483636
----------------+----------------------------------------------------------------
By using delta method this is WTP values from above coefficients

---------------------------------------------------------------------------------
choice | Coefficient Std. err. z P>|z| [95% conf. interval]
----------------+----------------------------------------------------------------
educ_3 | 81.35161 12.11413 6.72 0.000 57.60835 105.0949
educ_2 | 124.0298 15.3077 8.10 0.000 94.02728 154.0324
educ_1 | 151.8376 18.60096 8.16 0.000 115.3804 188.2948
housing_adv | 67.38103 10.34456 6.51 0.000 47.10607 87.65599
housing_basic | 53.77342 9.724813 5.53 0.000 34.71314 72.8337
district_center | 12.21744 7.400651 1.65 0.099 -2.287574 26.72245
zone_center | 31.03061 8.341431 3.72 0.000 14.68171 47.37952
delay_amnth | 2.504221 7.204097 0.35 0.728 -11.61555 16.62399
pay_ontime | 22.90385 7.880286 2.91 0.004 7.458774 38.34893
mgmt_support | 37.52427 6.865265 5.47 0.000 24.0686 50.97994
equipdrugs | 13.04514 5.259104 2.48 0.013 2.737485 23.35279
infrastructure | 18.26529 5.472677 3.34 0.001 7.539042 28.99154
---------------------------------------------------------------------------------

When I coded salary in money value WTP becomes this

choice	Coefficient	Std. err.	P>z	[95% conf. interval]
Education after 3 years	4521.14	641.81	0.0000	3263.209	5779.07
Education after 2 years	6898.89	808.39	0.0000	5314.473	8483.314
Education after 1 year	8381.96	973.80	0.0000	6473.338	10290.57
Provide advanced housing	3909.27	565.88	0.0000	2800.165	5018.381
Provide basic housing	3008.48	523.20	0.0000	1983.022	4033.931
Payment delays by a month	168.37	414.86	0.6850	-644.745	981.4947
Payment is on time	1353.39	438.91	0.0020	493.1342	2213.638
Location is at district center	1866.13	477.47	0.0000	930.3063	2801.962
Location is at zone center	780.12	419.44	0.0630	-41.97448	1602.21
Management is supportive	2129.70	405.45	0.0000	1335.023	2924.374
Adequate equipment and drugs	674.06	307.96	0.0290	70.45826	1277.652
Sufficient infrastructure	1163.40	311.96	0.0000	551.9651	1774.832

I am very confused and not able to understand which is the correct way.

[Hong Il Yoo]

ABD HILO: You're comparing apples to oranges. In the first case (salary increase as percentage), your WTP estimate for educ_3 suggests that "educ_3" gives as much extra utility as a 81.35% increase in salary. In the second case (salary in money), the corresponding estimate suggests that "educ_3" gives as much extra utility as a 4521.14 unit increase in salary (which would be a $45,211,400 increase if salary is measured in $10,000s).