Problem with Logistic Regression

[Jasmine Maddox]

Here is my analytic sample and variables code:

*1. Variables for analysis*
*Maternal Age: 5 year age groups*
tab mager, mis
sum mager, detail
gen mager_cat=1 if mager<20
replace mager_cat=2 if mager>=20&mager<=24
replace mager_cat=3 if mager>=25&mager<=29
replace mager_cat=4 if mager>=30&mager<=34
replace mager_cat=5 if mager>=35&mager<=39
replace mager_cat=6 if mager>=40&!missing(mager)
label var mager_cat "Mother's Age Categories"
label val mager_cat mager_c
label define mager_cat 1"1<20" 2"2:20-24" 3"3:25-29" 4"4:30-34" 5"5:35-39" 6"6:40+"

*Mother's Nativity*
tab mbstate_rec
gen nativity=1 if mbstate_rec==1
replace nativity=2 if mbstate_rec==2
replace nativity=3 if mbstate_rec==3
label var nativity "Citizenship"
label val nativity nativity
label define nativity 1"1:US Born" 2"2:Foreign Born" 3"3:Unknown"
tab nativity
drop if nativity==.

*Black ethnicity groups*
tab mrace15 mrace6
tab mrace31 mrace6
gen matrace=1 if (mrace6==2&!missing(mrace6)|mrace15==2&!missing(mr ace15)|mrace31==2&!missing(31))
tab matrace

*Marital Status*
tab dmar, mis
gen married=dmar
recode married 2=0
label var married "Marital status"
label val married yesno
label define yesno 1"1:Yes" 0"0:No"
tab married, mis


*Maternal Education: < HS, HS, Some college, College+*
tab meduc, mis
replace meduc=. if meduc==9
gen meduc_cat=1 if meduc<=2
replace meduc_cat=2 if meduc==3
replace meduc_cat=3 if meduc==4
replace meduc_cat=4 if meduc>=5&!missing(meduc)
label var meduc_cat "Mother's Education Categories"
label val meduc_cat meduc_cat
label define meduc_cat 1"1:Less than high school" 2"2:High School" 3"3:Some college" 4"4:College or more"
tab mrstate meduc if f_meduc==1,mis
tab dob_yy meduc if f_meduc==1, mis row

*Source of Payment: pay_rec*
tab pay_rec
rename pay pay_og
gen pay=pay_rec
replace pay=. if pay==9
label var pay "Payment"
label val pay pay
label define pay 1"1:Medicaid" 2"2:Private Insurance" 3"3:Self-pay" 4"4:Other"
tab dob_yy pay if f_meduc==1, mis
tab mrstate pay if f_meduc==1, mis

*WIC: Yes/No*
gen wicuse=wic
replace wicuse="2" if wicuse=="U"
replace wicuse="1" if wicuse=="Y"
replace wicuse="0" if wicuse=="N"
destring wicuse, replace
replace wicuse=. if wicuse==2
label var wicuse "Wic Use"
label val wicuse yesno
tab dob_yy wicuse if f_meduc==1, mis row

*Prenatal care: 1st trimester, 2nd tri, 3rd or No PNC; precare5 with 5 to missing*
replace precare5=. if precare5==5
gen precare_cat=1 if precare5==1c
replace precare_cat=2 if precare5>=2&precare5<=3
replace precare_cat=3 if precare5==4
label var precare_cat "Prenatal Care Initiation"
label val precare_cat precare_cat
label define precare_cat 1"1:Early" 2"2:Late" 3"3:None"
tab dob_yy precare_cat if f_meduc==1, mis row

/*Prenatal care any - Y/N*
gen pnc=1 if precare5>=1&precare5<=3
replace pnc=0 if precare5==4
label var pnc "Prenatal care"
label val pnc yesno
tab precare5, mis
tab pnc,mis*/

*Prepregnancy BMI: Underweight, Normal, Overweight, Obese*
replace bmi_r=. if bmi_r==9
gen bmi_cat4=1 if bmi_r==1
replace bmi_cat4=2 if bmi_r==2
replace bmi_cat4=3 if bmi_r==3
replace bmi_cat4=4 if bmi_r>=4&!missing(bmi_r)
label var bmi_cat4 "BMI Categories"
label val bmi_cat4 bmi_cat4
label define bmi_cat4 1"1:Underweight" 2"2:Normal" 3"3:Overweight" 4"4:Obese"
tab dob_yy bmi_cat4 if f_meduc==1, mis row

*Pre-pregnancy smoking: Yes/No -cig_0*
tab cig_0, mis
replace cig_0=. if cig_0==99
gen cigprepreg_cat2=0 if cig_0==0
replace cigprepreg_cat2=1 if cig_0>=1&!missing(cig_0)
label var cigprepreg_cat2 "Cigarette Use Prepregnancy"
label val cigprepreg_cat2 yesno
tab dob_yy cigprepreg_cat2 if f_meduc==1, mis row

*Pregnancy smoking: Any smoking during pregnancy - cig_rec*
gen ciguse=cig_rec
replace ciguse="2" if ciguse=="U"
replace ciguse="1" if ciguse=="Y"
replace ciguse="0" if ciguse=="N"
destring ciguse, replace
replace ciguse=. if ciguse==2
label var ciguse "Cigarette Use During Pregrnancy"
label val ciguse ciguse
label define ciguse 0"0:No" 1"1:Yes"
tab dob_yy ciguse if f_meduc==1, mis row

*Combined pregnancy smoking status:
tab cigprepreg_cat2 ciguse, mis
gen ciguse_tot=0 if cigprepreg_cat2==0&ciguse==0
replace ciguse_tot=1 if cigprepreg_cat2==1&ciguse==0
replace ciguse_tot=2 if cigprepreg_cat2==0&ciguse==1
replace ciguse_tot=3 if cigprepreg_cat2==1&ciguse==1
tab ciguse_tot, mis
label var ciguse_tot "Cig use before/during preg"
label val ciguse_tot cigusetot
label define cigusetot 0"0:Never" 1"1:before only" 2"2:during only" 3"3:before/during"

gen ciguse_tot3=ciguse_tot
recode ciguse_tot3 3=2
label var ciguse_tot3 "Cig use before/during preg"
label val ciguse_tot3 cigusetot3
label define cigusetot3 0"0:Never" 1"1:before only" 2"2:during and/or before"

/*Pre-diabetes Y/N*
tab rf_pdiab,mis
gen prediab=1 if rf_pdiab=="Y"
replace prediab=0 if rf_pdiab=="N"
replace prediab=. if rf_pdiab=="U"&!missing(rf_pdiab)
label var prediab "Pre-diabetes"
label val prediab yesno
tab rf_pdiab prediab,mis
tab dob_yy prediab if f_meduc==1, mis row
*/
/*Gestational diabetes Y/N*
tab rf_gdiab,mis
gen gestdiab=1 if rf_gdiab=="Y"
replace gestdiab=0 if rf_gdiab=="N"
replace gestdiab=. if rf_gdiab=="U"&!missing(rf_gdiab)
label var gestdiab "Gestational diabetes"
label val gestdiab yesno
tab rf_gdiab gestdiab,mis
tab dob_yy gestdiab if f_meduc==1, mis row
*/
/*Diabetes*
gen diabetes=1 if (prediab==1|gestdiab==1)
replace diabetes=0 if (prediab==0&gestdiab==0)
replace diabetes=. if (prediab==.&gestdiab==.)
label var diabetes "Diabetes"
label val diabetes yesno
tab diabetes,mis
*/

*Diabetes categories*
tab prediab gestdiab, missing
gen diabetescat=1 if prediab==0&gestdiab==0
replace diabetescat=2 if prediab==1&gestdiab==0
replace diabetescat=3 if prediab==0&gestdiab==1
tab diabetescat prediab, missing
tab diabetescat gestdiab, missing
label var diabetescat "Diabetes categories"
label val diabetescat diabcat
label define diabcat 1"1:None" 2"2:Prediab" 3"3:GDM"

*Hypertension: gestational, pre-pregnancy, none (use rf_phype, rf_ghype, rf_ehype)*
tab rf_phype rf_ghype, mis
tab rf_ehype rf_ghype, mis col
tab rf_ehype rf_phype, mis col
gen hyperten1=0 if rf_ghype=="N"&rf_phype=="N"
replace hyperten1=1 if rf_phype=="Y"&rf_ghype=="N"
replace hyperten1=2 if rf_phype=="N"&rf_ghype=="Y"
tab hyperten1, mis
tab hyperten1 rf_ehype, mis
gen hyperten=hyperten1
replace hyperten=2 if hyperten1==0&rf_ehype=="Y"
tab hyperten, mis
label var hyperten "Hypertension"
label val hyperten hyperten
label define hyperten 0"0:None" 1"1:Prepregnancy" 2"2:Gestational"

*Birth order: 1,2, 3, 4+; 1 or 2+*
replace lbo_rec=. if lbo_rec==9
gen lbo_cat=1 if lbo_rec==1
replace lbo_cat=2 if lbo_rec==2
replace lbo_cat=3 if lbo_rec==3
replace lbo_cat=4 if lbo_rec>=4&!missing(lbo_rec)
label var lbo_cat "Birth Order 1, 2, 3, 4+"
label val lbo_cat lbo_cat
label define lbo_cat 1"1:1" 2"2:2" 3"3:3" 4"4:4+"
tab dob_yy lbo_rec, mis row

*Gestational week at birth*
tab oegest_comb, mis
replace oegest_comb=. if oegest_comb==99
gen gestage=1 if oegest_comb<37
replace gestage=2 if (oegest_comb==37|oegest_comb==38)
replace gestage=3 if (oegest_comb==39|oegest_comb==40)
replace gestage=4 if oegest_comb==41
replace gestage=5 if oegest_comb>=42&!missing(oegest_comb)
tab oegest_comb gestage,mis
label var gestage "Gest Age Cat"
label val gestage gestage
label define gestage 1"1:<37" 2"2:37-38" 3"3: 39-40" 4"4:41" 5"5:42+"

*Infant Sex: M/F*
gen infsex=sex
replace infsex="0" if infsex=="M"
replace infsex="1" if infsex=="F"
destring infsex, replace
label var infsex "Infant Sex"
label val infsex infsex
label define infsex 0"0:Male" 1"1:Female"


*Birthweight*
tab dbwt
recode dbwt 9999=.
gen bw_cat=1 if dbwt<1500
replace bw_cat=2 if dbwt>=1500&dbwt<2500
replace bw_cat=3 if dbwt>=2500&dbwt<4000
replace bw_cat=4 if dbwt>=4000&!missing(dbwt)
label var bw_cat "Birthweight Cat"
label val bw_cat bw_cat
label define bw_cat 1"1:<1500" 2"2:1500-2499" 3"3:2500-3999" 4"4:4000+"

**Breastfeeding on discharge*
generate bfed_new=1 if bfed=="Y"
replace bfed_new=0 if bfed=="N"
replace bfed_new=. if bfed=="U"&!missing(bfed)
label var bfed_new "Breastfeeding on discharge"
label val bfed_new yesno
tab bfed_new,mis
tab dob_yy bfed_new if f_meduc==1, mis row
tab dob_yy bfed_new if f_meduc==1& ostate!="CA", mis row
tab dob_yy bfed_new if f_meduc==1& ostate!="MI", mis row
*dropped California and Michigan because they did not collect breastfeeding data for years analyzed*

/*Admission to NICU*
tab ab_nicu,mis
gen adm_nicu=1 if ab_nicu=="Y"
replace adm_nicu=0 if ab_nicu=="N"
replace adm_nicu=. if ab_nicu=="U"&!missing(ab_nicu)
label var adm_nicu "Admission to NICU"
label val adm_nicu yesno
tab ab_nicu adm_nicu,mis
tab dob_yy adm_nicu if f_meduc==1, mis row

*Infant transferred*
tab itran,mis
gen inftrans=1 if itran=="Y"
replace inftrans=0 if itran=="N"
replace inftrans=. if itran=="U"&!missing(itran)
label var inftrans "Infant transferred"
label val inftrans yesno
tab itran inftrans,mis
tab dob_yy inftrans if f_meduc==1, mis row

tab bfed_new itran, mis
*/

*NICU admission and/or infant transferred*
tab ab_nicu itran, mis
gen nicu=1 if ab_nicu=="Y"
replace nicu=1 if itran=="Y"
replace nicu=0 if (itran=="N"&ab_nicu=="N")|(ab_nicu=="U"&itran=="N" )|(ab_nicu=="N"&itran=="U")
tab nicu, mis
label var nicu "NICU Admission"
label val nicu nicu
label define nicu 0"0:No" 1"1:Yes"
tab dob_yy nicu if f_meduc==1, mis row


*2. Analytic sample: residents, Medicaid births, term births, all states reporting breastfeeding (excludes California/Michigan)*
tab restatus
drop if restatus==4

tab pay
keep if pay==1

tab bfed_new, mis
tab mrstate bfed_new, mis row
tab ostate bfed_new, mis row
drop if mrstate=="CA"
drop if mrstate=="MI"

Im trying to run this logistic regression:
logit gestage i.hyperten i.dob_yy i.precare_cat i.wicuse ib2.mager_cat ib2.meduc_cat ib1.lbo_cat ib2.bmi_cat4 i.ciguse_tot3 i.infsex i.diabetescat if nativity==1, or

But getting this error message:


outcome does not vary; remember:
0 = negative outcome,
all other nonmissing values = positive outcome
r(2000);

please help

[Mike Lacy]

The error message you received is relatively direct: The -logit- command requires that your outcome variable be coded with two values, 0 and not-zero. It appears that you have tried to code your gestage variable into the values 1 through 5. This makes your original variable oegest_comb into an ordered categorical variable. You could analyze it with an ordinal or multinomial logit model, but not the binary logit model you are trying to use. Or, you could code it into a 0/1 binary variable and use -logit-. However, while I'm not experienced in the substance of your research area, I would not recommend that, as collapsing an outcome or explanatory variable discards information. There are some situations in which that's a good idea, but not typically. Some other statistical approach would likely be preferable.

Beyond that, the code (syntax) you has various oddities. Here are two suggestions:

1) There are shorter and less error-prone ways to change the categorization of a variable than what you have done with a series of -replace- commands. Among the possibilities would be the -recode- command. For example, your mager_cat variable might be created like this:

recode mager (min/20 = 1) (21/24 = 2) (25/29 = 3) (30/34 = 4) (35/39 = 5) (40/max = 6) (missing = .)

However, again, I'm not recommending a categorization practice like this unless you have some good reason.

2) /* and */ will make everything in between them be ignored by Stata. The fact that you have left those in at several places in your code is a strange thing to do if you are presenting example code, which makes me wonder whether you know what these character pairs do. See -help comment-