Reference Category in Ordered Logit

[Tungga Dewi]

Dear all,

I'am using STATA 12 for mac. Currently, I am running an ordered legit regression to find out relationship between certain employment status and poverty. I have 6 levels employment status as my regressor (1=self employed, 2=self-employed with temporary workers, 3=self-employed with permanent worker, 4=employee, 5=casual worker, 6=family worker) and I want to choose one status as reference category. In this case, I want to choose "employee" as my reference category. How to create dummy variable for this?

gen self_employ=1 if employ_status=1
replace self_employ=0 if employ_status=4

Is that command right?. However, how to treat those missing values from other status of employment in that variable?. Will this create a problem in regression?

Any feedbacks or help from you is really appreciated!

Thanks

[Richard Williams]

See -help fvvarlist-. If employment status is an explanatory variable, just do something like

ologit poverty ib3.employ_status

If employment status is your dependent variable, it doesn't make any sense to refer to a reference category in ologit.

[Tungga Dewi]

Hi Richard,

Yes, it solved now!! Thank you SO MUCH for your answer

[Carlo Lazzaro]

Tungga:
if you want to solve missing values issue, you should check for their informativeness first.
You may want to take a look at -mi- and related entries in Stata .pdf manual.
You may also be interested in Paul Allison's http://www.sagepub.in/textbooks/Book9419 and in www.missingdata.org.uk/; the latter is maintained by Jeremy Bartlett, whose posts appear on this forum from time to time.
Stata applies listwise deletion for observations with missing values in any of the variables (dependent and/or independent): that approach boils down to reduce your sample size (with the subsequent statistical problems).

[Richard Williams]

I thought Tungga was referring to the missing values that would have been created with the originally proposed coding strategy. That shouldn't be a concern when using factor variables. But if the problem is indeed broader than what I thought then I second Carlo's advice. Those are good references.

[Marcos Almeida]

The theme is far from my field. That said, IMHO, it strikes me as "debatable" whether the outcome variable in #1 is really ordered.

Being this so, maybe a multinomial logistic regression under - mlogit - could provide an interesting solution.

By the way, under - mlogit - you may select the outcome baseline level, as you wished. You don't need to stick with the lower levels versus the higher levels, or vice-versa.

You may choose any level as "the" reference, but it's better to select among the most prevalent ones.

[Nahed Eddai]

Dear all,
How to specify the reference category for the dependent variable in the case of ordered logit in stata. When I use ib.dependent variable, I get the following message: depvar may not be a factor variable.
Best regards

[Carlo Lazzaro]

Nahed:
see Richard's helpful reply #5 in this thread.
In addition, quoting -ologit- entry in Stata .pdf manual (page 1833 for Stata 16):

ologit and oprobit begin by tabulating the dependent variable. Category i = 1 is defined as

the minimum value of the variable, i = 2 as the next ordered value, and so on, for the empirically

determined k categories.

[Nahed Eddai]

Thank you for your answer. So if I understood correctly stata takes by default the category 1 as reference in an ordered logit.
Best regards

[Carlo Lazzaro]

Nahed:
the following toy-example hopefully clarifies what happens behind the -ologit- curtain:

use "C:\Program Files\Stata16\ado\base\a\auto.dta"
. ologit rep78 i.foreign

Iteration 0:   log likelihood = -93.692061 
Iteration 1:   log likelihood = -79.696089 
Iteration 2:   log likelihood = -79.034005 
Iteration 3:   log likelihood = -79.029244 
Iteration 4:   log likelihood = -79.029243 

Ordered logistic regression                     Number of obs     =         69
                                                LR chi2(1)        =      29.33
                                                Prob > chi2       =     0.0000
Log likelihood = -79.029243                     Pseudo R2         =     0.1565

------------------------------------------------------------------------------
       rep78 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     foreign |
    Foreign  |    2.98155   .6203644     4.81   0.000     1.765658    4.197442
-------------+----------------------------------------------------------------
       /cut1 |  -3.158382   .7224269                     -4.574313   -1.742452
       /cut2 |  -1.362642   .3557343                     -2.059868   -.6654154
       /cut3 |   1.232161   .3431227                      .5596532     1.90467
       /cut4 |   3.246209   .5556657                      2.157124    4.335293
------------------------------------------------------------------------------

. mat list e(b)

e(b)[1,6]
         rep78:      rep78:          /:          /:          /:          /:
            0b.          1.                                               
       foreign     foreign        cut1        cut2        cut3        cut4
y1           0     2.98155  -3.1583824  -1.3626418   1.2321614   3.2462088

. mat list e(cat)

e(cat)[1,5]
    c1  c2  c3  c4  c5
r1   1   2   3   4   5

.

[Nahed Eddai]

Dear all,
What are the hypotheses to be tested before running an ordered logit regression?
Best regards.

[Carlo Lazzaro]

Nahed:
the trivial assumption to be tested is that the levels of your categorical variable (regressand) are actually ranked (bad; decent; good).
Conversely, if the levels of your regressand are, say, the colors of the traffic light (green; yellow; red), you should go -mlogit-.

[Nahed Eddai]

Thank you Carlo. Please could you clarify for me what is the difference betwwen odds ratios and marginal effects for an ordered logit regression.
Best regards

[Carlo Lazzaro]

Nahed:
your first question can be replied with the help of the usual toy-example:

. use "https://www.stata-press.com/data/r16/fullauto.dta"
(Automobile Models)

. ologit rep78 i.foreign, or
Iteration 0:   log likelihood = -93.692061
Iteration 1:   log likelihood = -79.696089
Iteration 2:   log likelihood = -79.034005
Iteration 3:   log likelihood = -79.029244
Iteration 4:   log likelihood = -79.029243

Ordered logistic regression                     Number of obs     =         69
                                                LR chi2(1)        =      29.33
                                                Prob > chi2       =     0.0000
Log likelihood = -79.029243                     Pseudo R2         =     0.1565

------------------------------------------------------------------------------
       rep78 | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     foreign |
    Foreign  |   19.71836   12.23257     4.81   0.000     5.845417    66.51596
-------------+----------------------------------------------------------------
       /cut1 |  -3.158382   .7224269                     -4.574313   -1.742452
       /cut2 |  -1.362642   .3557343                     -2.059868   -.6654154
       /cut3 |   1.232161   .3431227                      .5596532     1.90467
       /cut4 |   3.246209   .5556657                      2.157124    4.335293
------------------------------------------------------------------------------
Note: Estimates are transformed only in the first equation.

. ologit rep78 i.foreign

Iteration 0:   log likelihood = -93.692061
Iteration 1:   log likelihood = -79.696089
Iteration 2:   log likelihood = -79.034005
Iteration 3:   log likelihood = -79.029244
Iteration 4:   log likelihood = -79.029243

Ordered logistic regression                     Number of obs     =         69
                                                LR chi2(1)        =      29.33
                                                Prob > chi2       =     0.0000
Log likelihood = -79.029243                     Pseudo R2         =     0.1565

------------------------------------------------------------------------------
       rep78 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
     foreign |
    Foreign  |    2.98155   .6203644     4.81   0.000     1.765658    4.197442
-------------+----------------------------------------------------------------
       /cut1 |  -3.158382   .7224269                     -4.574313   -1.742452
       /cut2 |  -1.362642   .3557343                     -2.059868   -.6654154
       /cut3 |   1.232161   .3431227                      .5596532     1.90467
       /cut4 |   3.246209   .5556657                      2.157124    4.335293
------------------------------------------------------------------------------

. di exp(2.98155)
19.718356

.

As far as your (too broad) second question is concerned, please see -margins- entry in Stata .pdf manual. Thanks.

[Nahed Eddai]

Dear all,
How to fix a reference category in a binary logit for the dependent variable.
Best regards