How to change the reference category under Multinomial Logistic Regression using mi estimate?

[Mei Mei Liu]

Hello Stata experts,

Sorry for keeping asking questions again.

I wish to know how to change the reference category if it's under Multinomial Logistic Regression, using mi estimate.

I wish to change the reference category of variable, REGION. Region variable is my dependent variable (1=Northeast, 2= Midwest, 3=South, 4=West).

I have already tried, ib1.region or state char REGION[omit]1, but both commands don't work.

Wish any experts can give some comments~~

Thank you!!


************************************
Option 1.

. mi estimate,rrr:svy:mlogit ib1.REGION i.SEX
depvar may not be a factor variable
an error occurred when svy executed mlogit
an error occurred when mi estimate executed svy:mlogit on m=1

Option 2.

. char REGION[omit]1

. mi estimate,rrr:svy:mlogit REGION i.SEX

Multiple-imputation estimates Imputations = 5
Survey: Multinomial logistic regression Number of obs = 63510

Number of strata = 140 Population size = 2046598491
Number of PSUs = 1202
Average RVI = 0.0000
Largest FMI = 0.0000
Complete DF = 1062
DF adjustment: Small sample DF: min = 1060.01
avg = 1060.01
max = 1060.01
Model F test: Equal FMI F( 3, 1060.0) = 2.22
Within VCE type: Linearized Prob > F = 0.0839

------------------------------------------------------------------------------
REGION | RRR Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
Northeast |
SEX |
Female | .9285625 .0539682 -1.28 0.203 .8284812 1.040734
_cons | .4785517 .0624778 -5.65 0.000 .3704011 .6182803
-------------+----------------------------------------------------------------
Midwest |
SEX |
Female | 1.003805 .0523793 0.07 0.942 .906113 1.11203
_cons | .5617586 .0789383 -4.10 0.000 .4263852 .7401117
-------------+----------------------------------------------------------------
South | (base outcome)
-------------+----------------------------------------------------------------
West |
SEX |
Female | .9139301 .0361514 -2.28 0.023 .8456768 .987692
_cons | .5691837 .0701109 -4.58 0.000 .4469751 .7248056
------------------------------------------------------------------------------


************************************


My Stata command and ouput are as follows.

. tab REGION

Geographic |
region | Freq. Percent Cum.
------------+-----------------------------------
Northeast | 70,602 18.53 18.53
Midwest | 93,312 24.49 43.02
South | 120,792 31.70 74.71
West | 96,354 25.29 100.00
------------+-----------------------------------
Total | 381,060 100.00

. mi estimate,rrr:svy:mlogit REGION i.SEX

Multiple-imputation estimates Imputations = 5
Survey: Multinomial logistic regression Number of obs = 63510

Number of strata = 140 Population size = 2046598491
Number of PSUs = 1202
Average RVI = 0.0000
Largest FMI = 0.0000
Complete DF = 1062
DF adjustment: Small sample DF: min = 1060.01
avg = 1060.01
max = 1060.01
Model F test: Equal FMI F( 3, 1060.0) = 2.22
Within VCE type: Linearized Prob > F = 0.0839

------------------------------------------------------------------------------
REGION | RRR Std. Err. t P>|t| [95% Conf. Interval]
-------------+----------------------------------------------------------------
Northeast |
SEX |
Female | .9285625 .0539682 -1.28 0.203 .8284812 1.040734
_cons | .4785517 .0624778 -5.65 0.000 .3704011 .6182803
-------------+----------------------------------------------------------------
Midwest |
SEX |
Female | 1.003805 .0523793 0.07 0.942 .906113 1.11203
_cons | .5617586 .0789383 -4.10 0.000 .4263852 .7401117
-------------+----------------------------------------------------------------
South | (base outcome)
-------------+----------------------------------------------------------------
West |
SEX |
Female | .9139301 .0361514 -2.28 0.023 .8456768 .987692
_cons | .5691837 .0701109 -4.58 0.000 .4469751 .7248056
------------------------------------------------------------------------------

[Richard Williams]

with plain old mlogit (without all the mi and svy) it would be something like

mlogit REGION i.SEX, b(1)

where b is short for baseoutcome. Try that and see if it works.

Also, use code tags when posting your output -- that makes it easier to read. See pt. #12 of the FAQ.

[Mei Mei Liu]

Hello Richard,

Thank you so much!!

The command works!!!!

I will use the cod tags next time when posting the output!! Thank you thank you!!