Greetings,
I'm running Stata 15.1 on a Mac OS and working with cross-sectional survey data. My dependent variable is a 7-point measure of preferred immigration levels, the distribution of which is skewed towards 'increase' responses:
Code:
. tab increase_immig
Q68 | Freq. Percent Cum.
----------------------------+-----------------------------------
Decreased a lot | 49 4.89 4.89
Decreased a moderate amount | 49 4.89 9.77
Decreased a little | 68 6.78 16.55
Kept the same as now | 325 32.40 48.95
Increased a little | 186 18.54 67.50
Increased a moderate amount | 199 19.84 87.34
Increased a lot | 127 12.66 100.00
----------------------------+-----------------------------------
Total | 1,003 100.00
My question relates to how best to model this variable so that the results are presentable. I initially used ologit, but subsequently found that the model violated the proportional odds assumption. An alternative that relaxes this assumption is to use the gologit2 command. The problem is that, with 7 different response categories, the output is going to be very difficult to tabulate and present in a paper. Collapsing the DV to 3 categories would make it easier, but then 'increase' becomes the modal response and I'd lose a lot of information. I'm thus wondering whether I can justify transforming the DV to run from 0 to 1 and then modeling it with 'fracreg logit' (an alternative is to treat it as continuous and use OLS, but I'd rather avoid it). More generally, I'm wondering how any of you would approach this dilemma (alternative suggestions?). Thanks in advance for your insight!
0 Response to 7-point ordinal dependent variable: Is it okay to transform and model with fracreg logit?
Post a Comment