Libraries

## Data

ddi <- read_ipums_ddi("/Volumes/Jyoti/Stat 2 /PROJECT/nhis_00012.xml")
data <- read_ipums_micro(ddi)

## Use of data from IPUMS NHIS is subject to conditions including that users
## should cite the data appropriately. Use command `ipums_conditions()` for more
## details.

data<- haven::zap_labels(data)

names(data) <- tolower(gsub(pattern = "_",replacement =  "",x =  names(data)))

Prepare variables

data$age_cat<-Recode(data$age, recodes= "18:24='18-24'; 25:29='25-29'; 30:39='30-39'; 40:49='40-49'; 50:85='50 and above'; else=NA", as.factor=T)




#Age cut into intervals
data$ageinterval<-cut(data$age, breaks=c(18,29,39,59,79,99))




#depression level
data$depfeelevl<-as.factor(data$depfeelevl)
data$depfeelevl<- car::Recode(data$depfeelevl,
                       recodes="1='A Lot'; 2='A Little'; 
                       3='Between a Little and a Lot'; 7:9=NA; else=NA",
                      as.factor=T)
data$depfeelevl<- as.ordered(data$depfeelevl)

# medication for depression
data$deprx <- as.factor(data$deprx)

data$deprx<- car::Recode(data$deprx,
                       recodes="1=0; 2=1;else=NA",
                       as.factor=T)



#currently Pregnant 
data$pregnantnow<-as.factor(data$pregnantnow)
data$curpreg<-car::Recode(data$pregnantnow,
                          recodes="0='Yes';else=NA",
                          as.factor=T)
        
#education level

data$educ<-Recode(data$educ,
                        recodes="100:116 ='Less than HS'; 200:202='HS Diploma/GED'; 300:303='Some college';400= 'Undergraduate Degree'; 500:503:= 'Graduate Degree';else=NA", as.factor = T)
data$educ<-as.factor(data$educ)



#employment status

data$empstat<- car::Recode(data$empstat,
                       recodes="100='Employed'; 200='Unemployed';else=NA",
                       as.factor=T)


# income grouping

data$famtotinc_cat<-Recode(data$famtotinc, recodes = "0:49999='Less than 50k'; 50000:99999='50-100k';100000:149999='100-150k';150000:199999='150-200k';200000:250000='200-250k';else=NA", as.factor = T)
data$famtotinc_cat<-as.ordered(data$famtotinc)


##race
data$race<- car::Recode(data$racea,
                       recodes="100 ='White'; 200 ='African American'; 
                       400:590= 'Asian/Others'; else=NA", 
                       as.factor=T)


#race/ethnicity
data$black<- car::Recode(data$hisprace,
                       recodes="03=1; 99=NA; else=0")

data$white<- car::Recode(data$hisprace,
                       recodes="02=1; 99=NA; else=0")

data$other<- car::Recode(data$hisprace,
                      recodes="4:7=1; 99=NA; else=0")

data$hispanic<- car::Recode(data$hisprace,
                       recodes="01=1; 99=NA; else=0")

data$hisprace<- as.factor(data$hisprace)

data$race_eth<-car::Recode(data$hisprace,
recodes="01='Hispanic'; 02='NH_White'; 03='NH_Black';04:07='NH_Other'; else=NA",
as.factor = T)
data$race_eth<-relevel(data$race_eth,
                          ref = "NH_White")

## marital status
data$mars<- car::Recode(data$marstat, 
                        recodes ="10:13='Married'; 20='Widowed'; 30:40='Divorced/Separated'; 
                        ; 50='Never Married'; else=NA", 
                        as.factor=T)

Filter

## Filter data
data<-data%>%
  filter(is.na(curpreg)==F)
data<-data%>%
  filter(is.na(educ)==F)
data<-data%>%
  filter(is.na(deprx)==F)
data<-data%>%
  filter(is.na(depfeelevl)==F)
data<-data%>%
  filter(is.na(empstat)==F)
data<-data%>%
  filter(is.na(race)==F)
data<-data%>%
  filter(is.na(marstat)==F)
data<-data%>%
  filter(is.na(race_eth)==F)
data<-data%>%
  filter(is.na(age_cat)==F)

options(survey.lonely.psu = "adjust")

des<-svydesign(ids=~1, strata=~strata, weights=~sampweight, data = data, , nest=T )
des

## Stratified Independent Sampling design (with replacement)
## svydesign(ids = ~1, strata = ~strata, weights = ~sampweight, 
##     data = data, , nest = T)

## Table 1: Demographic properties of pregnant women with depression (who are either taking or not taking prescription medication for depression)
label(data$depfeelevl) <- "Depression Level"
label(data$deprx) <- "Medication for Depression"
label(data$pregnantnow) <-"Currently Pregnant"
label(data$educ) <- "Education Level"
label(data$empstat)<- "Employment Status"
label(data$race_eth)<- "Race/Ethnicity"
label(data$mars)<- "Marital Status"
label(data$age_cat)<- "Age"

I ran my outcome variable with all my predictor variables to see if they are significant. I will then nest them.

fit1 <- svyglm(deprx~ educ+age_cat+race_eth+empstat+mars, design=des, family=binomial(link="logit"))

## Warning in eval(family$initialize): non-integer #successes in a binomial glm!

# Take a look at the output
gtsummary::tbl_regression(fit1, exp = TRUE) 

Characteristic	OR1	95% CI1	p-value
educ
Graduate Degree	—	—
HS Diploma/GED	1.09	0.95, 1.25	0.2
Less than HS	1.05	0.87, 1.27	0.6
Some college	1.20	1.05, 1.37	0.008
Undergraduate Degree	0.97	0.85, 1.12	0.7
age_cat
18-24	—	—
25-29	1.42	0.96, 2.12	0.082
30-39	1.37	0.98, 1.91	0.065
40-49	1.93	1.38, 2.70	<0.001
50 and above	2.46	1.81, 3.35	<0.001
race_eth
NH_White	—	—
Hispanic	0.67	0.54, 0.83	<0.001
NH_Black	0.64	0.55, 0.76	<0.001
NH_Other	0.37	0.27, 0.49	<0.001
empstat
Employed	—	—
Unemployed	1.99	1.80, 2.19	<0.001
mars
Divorced/Separated	—	—
Married	0.75	0.68, 0.83	<0.001
Never Married	0.76	0.66, 0.88	<0.001
Widowed	0.80	0.69, 0.92	0.001
1 OR = Odds Ratio, CI = Confidence Interval

## Survey design #First we tell R our survey design

options(survey.lonely.psu = "adjust")

library(dplyr)
sub<-data%>%
  select(depfeelevl, age_cat, curpreg, educ, deprx, empstat, race_eth, mars,depfeelevl, sampweight,strata) %>%
  filter( complete.cases(.))

options(survey.lonely.psu = "adjust")
des<-svydesign(ids=~1,
               strata=~strata,
               weights=~sampweight,
               data =sub)

Descriptive Analysis

table<-table1(~ educ + empstat + race_eth + mars + depfeelevl + age_cat | deprx, data=sub)
table

	0 (N=14268)	1 (N=4863)	Overall (N=19131)
Education Level
Graduate Degree	2128 (14.9%)	643 (13.2%)	2771 (14.5%)
HS Diploma/GED	3502 (24.5%)	1307 (26.9%)	4809 (25.1%)
Less than HS	1140 (8.0%)	448 (9.2%)	1588 (8.3%)
Some college	4214 (29.5%)	1570 (32.3%)	5784 (30.2%)
Undergraduate Degree	3284 (23.0%)	895 (18.4%)	4179 (21.8%)
Employment Status
Employed	7670 (53.8%)	1676 (34.5%)	9346 (48.9%)
Unemployed	6598 (46.2%)	3187 (65.5%)	9785 (51.1%)
Race/Ethnicity
NH_White	11248 (78.8%)	4197 (86.3%)	15445 (80.7%)
Hispanic	847 (5.9%)	224 (4.6%)	1071 (5.6%)
NH_Black	1352 (9.5%)	350 (7.2%)	1702 (8.9%)
NH_Other	821 (5.8%)	92 (1.9%)	913 (4.8%)
Marital Status
Divorced/Separated	2834 (19.9%)	1280 (26.3%)	4114 (21.5%)
Married	6144 (43.1%)	1961 (40.3%)	8105 (42.4%)
Never Married	3295 (23.1%)	772 (15.9%)	4067 (21.3%)
Widowed	1995 (14.0%)	850 (17.5%)	2845 (14.9%)
Depression Level
A Little	1123 (7.9%)	1176 (24.2%)	2299 (12.0%)
A Lot	8507 (59.6%)	1666 (34.3%)	10173 (53.2%)
Between a Little and a Lot	4638 (32.5%)	2021 (41.6%)	6659 (34.8%)
Age
18-24	620 (4.3%)	78 (1.6%)	698 (3.6%)
25-29	729 (5.1%)	105 (2.2%)	834 (4.4%)
30-39	1513 (10.6%)	248 (5.1%)	1761 (9.2%)
40-49	1240 (8.7%)	264 (5.4%)	1504 (7.9%)
50 and above	10166 (71.3%)	4168 (85.7%)	14334 (74.9%)

Logit Regression

#Logit model
fit.logit<-svyglm(deprx~educ+race_eth+empstat+mars,
                  design= des,
                  family=binomial)

## Warning in eval(family$initialize): non-integer #successes in a binomial glm!

summary(fit.logit)

## 
## Call:
## svyglm(formula = deprx ~ educ + race_eth + empstat + mars, design = des, 
##     family = binomial)
## 
## Survey design:
## svydesign(ids = ~1, strata = ~strata, weights = ~sampweight, 
##     data = sub)
## 
## Coefficients:
##                          Estimate Std. Error t value Pr(>|t|)    
## (Intercept)              -1.22252    0.07092 -17.237  < 2e-16 ***
## educHS Diploma/GED        0.05230    0.06889   0.759  0.44776    
## educLess than HS          0.02342    0.09534   0.246  0.80595    
## educSome college          0.13679    0.06733   2.031  0.04222 *  
## educUndergraduate Degree -0.05155    0.07188  -0.717  0.47329    
## race_ethHispanic         -0.44134    0.10704  -4.123 3.75e-05 ***
## race_ethNH_Black         -0.42271    0.08159  -5.181 2.23e-07 ***
## race_ethNH_Other         -1.04345    0.14642  -7.127 1.07e-12 ***
## empstatUnemployed         0.80749    0.04605  17.534  < 2e-16 ***
## marsMarried              -0.31549    0.05222  -6.042 1.55e-09 ***
## marsNever Married        -0.64373    0.07065  -9.112  < 2e-16 ***
## marsWidowed              -0.20725    0.07079  -2.928  0.00342 ** 
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## (Dispersion parameter for binomial family taken to be 0.9959921)
## 
## Number of Fisher Scoring iterations: 4

Here, we only see coefficients, the direction of relationship, and the significance. Now, we need to calculate odds ratios and confidence intervals.

Get odd ratios and confidence intervals for the estimates

library(gtsummary)
fit.logit%>%
  tbl_regression(exponentiate=TRUE )

Characteristic	OR1	95% CI1	p-value
Education Level
Graduate Degree	—	—
HS Diploma/GED	1.05	0.92, 1.21	0.4
Less than HS	1.02	0.85, 1.23	0.8
Some college	1.15	1.00, 1.31	0.042
Undergraduate Degree	0.95	0.82, 1.09	0.5
Race/Ethnicity
NH_White	—	—
Hispanic	0.64	0.52, 0.79	<0.001
NH_Black	0.66	0.56, 0.77	<0.001
NH_Other	0.35	0.26, 0.47	<0.001
Employment Status
Employed	—	—
Unemployed	2.24	2.05, 2.45	<0.001
Marital Status
Divorced/Separated	—	—
Married	0.73	0.66, 0.81	<0.001
Never Married	0.53	0.46, 0.60	<0.001
Widowed	0.81	0.71, 0.93	0.003
1 OR = Odds Ratio, CI = Confidence Interval

library(sjPlot)

## #refugeeswelcome

plot_model(fit.logit,
           axis.lim = c(.1, 10),
           title = " Figure 1. Odds Ratios for Depression")

The figure above shows that women with some college education have higher odds of taking depression medication. In case of race and ethnicity, Whites have higher odds of taking depression medication compared to other groups. In terms of employment status, unemployed women are more likely to take depression medication. Finally, divorced/separated have higher odds of taking depression medication.

knitr::kable(data.frame(OR = exp(coef(fit.logit)), ci = exp(confint(fit.logit)))) 

	OR	ci.2.5..	ci.97.5..
(Intercept)	0.2944862	0.2562664	0.3384061
educHS Diploma/GED	1.0536946	0.9205934	1.2060399
educLess than HS	1.0236982	0.8491989	1.2340550
educSome college	1.1465826	1.0048191	1.3083466
educUndergraduate Degree	0.9497590	0.8249500	1.0934506
race_ethHispanic	0.6431729	0.5214499	0.7933098
race_ethNH_Black	0.6552702	0.5584242	0.7689120
race_ethNH_Other	0.3522372	0.2643607	0.4693247
empstatUnemployed	2.2422786	2.0487447	2.4540946
marsMarried	0.7294344	0.6584710	0.8080455
marsNever Married	0.5253274	0.4573956	0.6033484
marsWidowed	0.8128203	0.7075174	0.9337959

Fitted values

#get a series of predicted probabilities for different "types" of people for each model
#ref_grid will generate all possible combinations of predictors from a model

library(emmeans)
rg<-ref_grid(fit.logit)

marg_logit<-emmeans(object = rg,
              specs = c( "educ",  "race_eth", "empstat", "mars"),
              type="response" )

knitr::kable(marg_logit,  digits = 4)

educ	race_eth	empstat	mars	prob	SE	df	asymp.LCL	asymp.UCL
Graduate Degree	NH_White	Employed	Divorced/Separated	0.2275	0.0125	Inf	0.2040	0.2528
HS Diploma/GED	NH_White	Employed	Divorced/Separated	0.2368	0.0114	Inf	0.2152	0.2599
Less than HS	NH_White	Employed	Divorced/Separated	0.2316	0.0163	Inf	0.2013	0.2650
Some college	NH_White	Employed	Divorced/Separated	0.2524	0.0111	Inf	0.2312	0.2749
Undergraduate Degree	NH_White	Employed	Divorced/Separated	0.2186	0.0111	Inf	0.1976	0.2410
Graduate Degree	Hispanic	Employed	Divorced/Separated	0.1592	0.0166	Inf	0.1293	0.1945
HS Diploma/GED	Hispanic	Employed	Divorced/Separated	0.1664	0.0163	Inf	0.1368	0.2009
Less than HS	Hispanic	Employed	Divorced/Separated	0.1624	0.0163	Inf	0.1329	0.1969
Some college	Hispanic	Employed	Divorced/Separated	0.1784	0.0173	Inf	0.1469	0.2149
Undergraduate Degree	Hispanic	Employed	Divorced/Separated	0.1525	0.0154	Inf	0.1246	0.1853
Graduate Degree	NH_Black	Employed	Divorced/Separated	0.1618	0.0138	Inf	0.1365	0.1907
HS Diploma/GED	NH_Black	Employed	Divorced/Separated	0.1690	0.0132	Inf	0.1447	0.1964
Less than HS	NH_Black	Employed	Divorced/Separated	0.1650	0.0162	Inf	0.1356	0.1993
Some college	NH_Black	Employed	Divorced/Separated	0.1812	0.0135	Inf	0.1562	0.2092
Undergraduate Degree	NH_Black	Employed	Divorced/Separated	0.1549	0.0128	Inf	0.1314	0.1817
Graduate Degree	NH_Other	Employed	Divorced/Separated	0.0940	0.0132	Inf	0.0711	0.1233
HS Diploma/GED	NH_Other	Employed	Divorced/Separated	0.0985	0.0138	Inf	0.0746	0.1290
Less than HS	NH_Other	Employed	Divorced/Separated	0.0960	0.0144	Inf	0.0712	0.1282
Some college	NH_Other	Employed	Divorced/Separated	0.1063	0.0149	Inf	0.0805	0.1392
Undergraduate Degree	NH_Other	Employed	Divorced/Separated	0.0897	0.0127	Inf	0.0678	0.1178
Graduate Degree	NH_White	Unemployed	Divorced/Separated	0.3977	0.0173	Inf	0.3644	0.4320
HS Diploma/GED	NH_White	Unemployed	Divorced/Separated	0.4103	0.0141	Inf	0.3830	0.4382
Less than HS	NH_White	Unemployed	Divorced/Separated	0.4033	0.0205	Inf	0.3639	0.4441
Some college	NH_White	Unemployed	Divorced/Separated	0.4309	0.0138	Inf	0.4041	0.4581
Undergraduate Degree	NH_White	Unemployed	Divorced/Separated	0.3854	0.0155	Inf	0.3555	0.4162
Graduate Degree	Hispanic	Unemployed	Divorced/Separated	0.2981	0.0263	Inf	0.2493	0.3520
HS Diploma/GED	Hispanic	Unemployed	Divorced/Separated	0.3092	0.0248	Inf	0.2627	0.3598
Less than HS	Hispanic	Unemployed	Divorced/Separated	0.3030	0.0245	Inf	0.2572	0.3531
Some college	Hispanic	Unemployed	Divorced/Separated	0.3275	0.0259	Inf	0.2788	0.3802
Undergraduate Degree	Hispanic	Unemployed	Divorced/Separated	0.2874	0.0247	Inf	0.2415	0.3382
Graduate Degree	NH_Black	Unemployed	Divorced/Separated	0.3020	0.0217	Inf	0.2613	0.3461
HS Diploma/GED	NH_Black	Unemployed	Divorced/Separated	0.3131	0.0195	Inf	0.2762	0.3527
Less than HS	NH_Black	Unemployed	Divorced/Separated	0.3070	0.0241	Inf	0.2619	0.3561
Some college	NH_Black	Unemployed	Divorced/Separated	0.3316	0.0198	Inf	0.2939	0.3716
Undergraduate Degree	NH_Black	Unemployed	Divorced/Separated	0.2913	0.0203	Inf	0.2531	0.3327
Graduate Degree	NH_Other	Unemployed	Divorced/Separated	0.1887	0.0242	Inf	0.1458	0.2407
HS Diploma/GED	NH_Other	Unemployed	Divorced/Separated	0.1968	0.0246	Inf	0.1531	0.2494
Less than HS	NH_Other	Unemployed	Divorced/Separated	0.1923	0.0256	Inf	0.1470	0.2475
Some college	NH_Other	Unemployed	Divorced/Separated	0.2105	0.0262	Inf	0.1638	0.2664
Undergraduate Degree	NH_Other	Unemployed	Divorced/Separated	0.1809	0.0233	Inf	0.1396	0.2312
Graduate Degree	NH_White	Employed	Married	0.1768	0.0090	Inf	0.1599	0.1952
HS Diploma/GED	NH_White	Employed	Married	0.1846	0.0085	Inf	0.1685	0.2018
Less than HS	NH_White	Employed	Married	0.1803	0.0132	Inf	0.1558	0.2076
Some college	NH_White	Employed	Married	0.1976	0.0084	Inf	0.1818	0.2145
Undergraduate Degree	NH_White	Employed	Married	0.1694	0.0079	Inf	0.1545	0.1856
Graduate Degree	Hispanic	Employed	Married	0.1214	0.0127	Inf	0.0985	0.1487
HS Diploma/GED	Hispanic	Employed	Married	0.1271	0.0128	Inf	0.1041	0.1542
Less than HS	Hispanic	Employed	Married	0.1239	0.0129	Inf	0.1007	0.1515
Some college	Hispanic	Employed	Married	0.1367	0.0136	Inf	0.1121	0.1658
Undergraduate Degree	Hispanic	Employed	Married	0.1160	0.0119	Inf	0.0947	0.1413
Graduate Degree	NH_Black	Employed	Married	0.1234	0.0108	Inf	0.1038	0.1461
HS Diploma/GED	NH_Black	Employed	Married	0.1292	0.0105	Inf	0.1099	0.1512
Less than HS	NH_Black	Employed	Married	0.1259	0.0131	Inf	0.1024	0.1540
Some college	NH_Black	Employed	Married	0.1390	0.0109	Inf	0.1189	0.1618
Undergraduate Degree	NH_Black	Employed	Married	0.1179	0.0100	Inf	0.0997	0.1390
Graduate Degree	NH_Other	Employed	Married	0.0703	0.0098	Inf	0.0534	0.0922
HS Diploma/GED	NH_Other	Employed	Married	0.0738	0.0104	Inf	0.0559	0.0969
Less than HS	NH_Other	Employed	Married	0.0719	0.0109	Inf	0.0532	0.0965
Some college	NH_Other	Employed	Married	0.0798	0.0112	Inf	0.0604	0.1048
Undergraduate Degree	NH_Other	Employed	Married	0.0670	0.0094	Inf	0.0508	0.0880
Graduate Degree	NH_White	Unemployed	Married	0.3251	0.0138	Inf	0.2987	0.3526
HS Diploma/GED	NH_White	Unemployed	Married	0.3367	0.0113	Inf	0.3150	0.3591
Less than HS	NH_White	Unemployed	Married	0.3302	0.0183	Inf	0.2954	0.3671
Some college	NH_White	Unemployed	Married	0.3558	0.0112	Inf	0.3341	0.3780
Undergraduate Degree	NH_White	Unemployed	Married	0.3139	0.0122	Inf	0.2905	0.3382
Graduate Degree	Hispanic	Unemployed	Married	0.2365	0.0218	Inf	0.1964	0.2819
HS Diploma/GED	Hispanic	Unemployed	Married	0.2461	0.0210	Inf	0.2074	0.2894
Less than HS	Hispanic	Unemployed	Married	0.2408	0.0210	Inf	0.2021	0.2843
Some college	Hispanic	Unemployed	Married	0.2621	0.0222	Inf	0.2210	0.3079
Undergraduate Degree	Hispanic	Unemployed	Married	0.2273	0.0205	Inf	0.1897	0.2699
Graduate Degree	NH_Black	Unemployed	Married	0.2399	0.0183	Inf	0.2059	0.2775
HS Diploma/GED	NH_Black	Unemployed	Married	0.2496	0.0169	Inf	0.2180	0.2841
Less than HS	NH_Black	Unemployed	Married	0.2442	0.0212	Inf	0.2051	0.2880
Some college	NH_Black	Unemployed	Married	0.2657	0.0174	Inf	0.2330	0.3012
Undergraduate Degree	NH_Black	Unemployed	Married	0.2306	0.0171	Inf	0.1988	0.2659
Graduate Degree	NH_Other	Unemployed	Married	0.1450	0.0189	Inf	0.1118	0.1862
HS Diploma/GED	NH_Other	Unemployed	Married	0.1517	0.0195	Inf	0.1173	0.1939
Less than HS	NH_Other	Unemployed	Married	0.1480	0.0205	Inf	0.1122	0.1927
Some college	NH_Other	Unemployed	Married	0.1628	0.0210	Inf	0.1258	0.2082
Undergraduate Degree	NH_Other	Unemployed	Married	0.1388	0.0182	Inf	0.1067	0.1785
Graduate Degree	NH_White	Employed	Never Married	0.1340	0.0092	Inf	0.1170	0.1529
HS Diploma/GED	NH_White	Employed	Never Married	0.1402	0.0084	Inf	0.1246	0.1573
Less than HS	NH_White	Employed	Never Married	0.1367	0.0117	Inf	0.1153	0.1614
Some college	NH_White	Employed	Never Married	0.1507	0.0088	Inf	0.1342	0.1687
Undergraduate Degree	NH_White	Employed	Never Married	0.1281	0.0079	Inf	0.1134	0.1444
Graduate Degree	Hispanic	Employed	Never Married	0.0905	0.0108	Inf	0.0714	0.1141
HS Diploma/GED	Hispanic	Employed	Never Married	0.0949	0.0107	Inf	0.0759	0.1180
Less than HS	Hispanic	Employed	Never Married	0.0924	0.0108	Inf	0.0733	0.1160
Some college	Hispanic	Employed	Never Married	0.1024	0.0116	Inf	0.0818	0.1275
Undergraduate Degree	Hispanic	Employed	Never Married	0.0863	0.0099	Inf	0.0688	0.1079
Graduate Degree	NH_Black	Employed	Never Married	0.0920	0.0091	Inf	0.0756	0.1116
HS Diploma/GED	NH_Black	Employed	Never Married	0.0965	0.0087	Inf	0.0807	0.1151
Less than HS	NH_Black	Employed	Never Married	0.0940	0.0107	Inf	0.0750	0.1172
Some college	NH_Black	Employed	Never Married	0.1041	0.0093	Inf	0.0872	0.1238
Undergraduate Degree	NH_Black	Employed	Never Married	0.0878	0.0083	Inf	0.0728	0.1056
Graduate Degree	NH_Other	Employed	Never Married	0.0517	0.0077	Inf	0.0385	0.0691
HS Diploma/GED	NH_Other	Employed	Never Married	0.0543	0.0080	Inf	0.0405	0.0724
Less than HS	NH_Other	Employed	Never Married	0.0528	0.0085	Inf	0.0385	0.0721
Some college	NH_Other	Employed	Never Married	0.0588	0.0088	Inf	0.0437	0.0786
Undergraduate Degree	NH_Other	Employed	Never Married	0.0492	0.0073	Inf	0.0367	0.0657
Graduate Degree	NH_White	Unemployed	Never Married	0.2575	0.0160	Inf	0.2275	0.2900
HS Diploma/GED	NH_White	Unemployed	Never Married	0.2677	0.0136	Inf	0.2419	0.2951
Less than HS	NH_White	Unemployed	Never Married	0.2620	0.0187	Inf	0.2271	0.3003
Some college	NH_White	Unemployed	Never Married	0.2846	0.0144	Inf	0.2573	0.3135
Undergraduate Degree	NH_White	Unemployed	Never Married	0.2478	0.0140	Inf	0.2214	0.2763
Graduate Degree	Hispanic	Unemployed	Never Married	0.1824	0.0202	Inf	0.1461	0.2253
HS Diploma/GED	Hispanic	Unemployed	Never Married	0.1903	0.0193	Inf	0.1554	0.2310
Less than HS	Hispanic	Unemployed	Never Married	0.1859	0.0194	Inf	0.1509	0.2269
Some college	Hispanic	Unemployed	Never Married	0.2037	0.0208	Inf	0.1660	0.2475
Undergraduate Degree	Hispanic	Unemployed	Never Married	0.1748	0.0186	Inf	0.1412	0.2144
Graduate Degree	NH_Black	Unemployed	Never Married	0.1852	0.0170	Inf	0.1541	0.2210
HS Diploma/GED	NH_Black	Unemployed	Never Married	0.1932	0.0156	Inf	0.1644	0.2257
Less than HS	NH_Black	Unemployed	Never Married	0.1888	0.0190	Inf	0.1544	0.2288
Some college	NH_Black	Unemployed	Never Married	0.2067	0.0166	Inf	0.1761	0.2412
Undergraduate Degree	NH_Black	Unemployed	Never Married	0.1776	0.0157	Inf	0.1489	0.2103
Graduate Degree	NH_Other	Unemployed	Never Married	0.1089	0.0157	Inf	0.0817	0.1437
HS Diploma/GED	NH_Other	Unemployed	Never Married	0.1141	0.0160	Inf	0.0862	0.1494
Less than HS	NH_Other	Unemployed	Never Married	0.1112	0.0168	Inf	0.0823	0.1485
Some college	NH_Other	Unemployed	Never Married	0.1229	0.0174	Inf	0.0926	0.1613
Undergraduate Degree	NH_Other	Unemployed	Never Married	0.1040	0.0149	Inf	0.0781	0.1371
Graduate Degree	NH_White	Employed	Widowed	0.1931	0.0133	Inf	0.1684	0.2206
HS Diploma/GED	NH_White	Employed	Widowed	0.2014	0.0125	Inf	0.1781	0.2270
Less than HS	NH_White	Employed	Widowed	0.1968	0.0157	Inf	0.1678	0.2295
Some college	NH_White	Employed	Widowed	0.2153	0.0128	Inf	0.1914	0.2414
Undergraduate Degree	NH_White	Employed	Widowed	0.1852	0.0121	Inf	0.1626	0.2102
Graduate Degree	Hispanic	Employed	Widowed	0.1334	0.0160	Inf	0.1051	0.1679
HS Diploma/GED	Hispanic	Employed	Widowed	0.1396	0.0158	Inf	0.1114	0.1735
Less than HS	Hispanic	Employed	Widowed	0.1361	0.0155	Inf	0.1086	0.1694
Some college	Hispanic	Employed	Widowed	0.1500	0.0169	Inf	0.1198	0.1863
Undergraduate Degree	Hispanic	Employed	Widowed	0.1276	0.0149	Inf	0.1011	0.1597
Graduate Degree	NH_Black	Employed	Widowed	0.1356	0.0135	Inf	0.1112	0.1643
HS Diploma/GED	NH_Black	Employed	Widowed	0.1418	0.0130	Inf	0.1181	0.1694
Less than HS	NH_Black	Employed	Widowed	0.1384	0.0151	Inf	0.1113	0.1707
Some college	NH_Black	Employed	Widowed	0.1524	0.0136	Inf	0.1276	0.1811
Undergraduate Degree	NH_Black	Employed	Widowed	0.1297	0.0126	Inf	0.1069	0.1564
Graduate Degree	NH_Other	Employed	Widowed	0.0778	0.0118	Inf	0.0575	0.1044
HS Diploma/GED	NH_Other	Employed	Widowed	0.0816	0.0123	Inf	0.0605	0.1092
Less than HS	NH_Other	Employed	Widowed	0.0795	0.0127	Inf	0.0579	0.1081
Some college	NH_Other	Employed	Widowed	0.0882	0.0134	Inf	0.0652	0.1181
Undergraduate Degree	NH_Other	Employed	Widowed	0.0741	0.0113	Inf	0.0548	0.0996
Graduate Degree	NH_White	Unemployed	Widowed	0.3493	0.0178	Inf	0.3152	0.3849
HS Diploma/GED	NH_White	Unemployed	Widowed	0.3612	0.0147	Inf	0.3329	0.3906
Less than HS	NH_White	Unemployed	Widowed	0.3546	0.0198	Inf	0.3168	0.3943
Some college	NH_White	Unemployed	Widowed	0.3810	0.0150	Inf	0.3520	0.4107
Undergraduate Degree	NH_White	Unemployed	Widowed	0.3376	0.0161	Inf	0.3068	0.3700
Graduate Degree	Hispanic	Unemployed	Widowed	0.2566	0.0256	Inf	0.2096	0.3100
HS Diploma/GED	Hispanic	Unemployed	Widowed	0.2667	0.0244	Inf	0.2217	0.3172
Less than HS	Hispanic	Unemployed	Widowed	0.2611	0.0237	Inf	0.2174	0.3101
Some college	Hispanic	Unemployed	Widowed	0.2836	0.0258	Inf	0.2359	0.3367
Undergraduate Degree	Hispanic	Unemployed	Widowed	0.2469	0.0242	Inf	0.2027	0.2972
Graduate Degree	NH_Black	Unemployed	Widowed	0.2602	0.0212	Inf	0.2208	0.3039
HS Diploma/GED	NH_Black	Unemployed	Widowed	0.2704	0.0193	Inf	0.2342	0.3099
Less than HS	NH_Black	Unemployed	Widowed	0.2647	0.0228	Inf	0.2226	0.3117
Some college	NH_Black	Unemployed	Widowed	0.2874	0.0200	Inf	0.2498	0.3281
Undergraduate Degree	NH_Black	Unemployed	Widowed	0.2504	0.0199	Inf	0.2134	0.2914
Graduate Degree	NH_Other	Unemployed	Widowed	0.1590	0.0219	Inf	0.1207	0.2067
HS Diploma/GED	NH_Other	Unemployed	Widowed	0.1661	0.0222	Inf	0.1270	0.2143
Less than HS	NH_Other	Unemployed	Widowed	0.1622	0.0228	Inf	0.1222	0.2120
Some college	NH_Other	Unemployed	Widowed	0.1781	0.0239	Inf	0.1360	0.2298
Undergraduate Degree	NH_Other	Unemployed	Widowed	0.1522	0.0210	Inf	0.1154	0.1982

Nested model comparison

Since the author is interested in how SES (Marital status, employment status, and race and ethnicity) mediates the effects of educational differences on prenatal depression on women, the author used nested model comparison.

Predictor variables are entered into the model in “blocks”, for example, in the study’s outcome variable, prenatal depression and instead of entering all variables in the model simultaneously, the author begins with the effect of educational attainment, then add the effect of other socio-demographic variables such as race and ethnicity, employment and marital status.

fit.logit1<-svyglm(I(deprx==1)~educ, design=des, family=binomial) #educational attainment  only

## Warning in eval(family$initialize): non-integer #successes in a binomial glm!

fit.logit2<-svyglm(I(deprx==1)~educ +race_eth+ empstat, design=des, family=binomial) #educational attainment only +race/ethnicity+employment status 

## Warning in eval(family$initialize): non-integer #successes in a binomial glm!

fit.logit3<-svyglm(I(deprx==1)~educ +race_eth++empstat+mars, design=des, family=binomial) #educational attainment only +race/ethnicity+employment status+ marital status

## Warning in eval(family$initialize): non-integer #successes in a binomial glm!

In model 1 we see women with some college education have higher odds (0.22) of taking medication for depression, compared to women with graduate degree, while women with undergraduate degree show the lower odds (-0.09) of taking medication for depression.

summary(fit.logit1)

## 
## Call:
## svyglm(formula = I(deprx == 1) ~ educ, design = des, family = binomial)
## 
## Survey design:
## svydesign(ids = ~1, strata = ~strata, weights = ~sampweight, 
##     data = sub)
## 
## Coefficients:
##                          Estimate Std. Error t value Pr(>|t|)    
## (Intercept)              -1.32345    0.05360 -24.690  < 2e-16 ***
## educHS Diploma/GED        0.19199    0.06675   2.876 0.004030 ** 
## educLess than HS          0.22090    0.08989   2.457 0.014005 *  
## educSome college          0.22449    0.06533   3.436 0.000591 ***
## educUndergraduate Degree -0.09690    0.07035  -1.377 0.168423    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## (Dispersion parameter for binomial family taken to be 1.000052)
## 
## Number of Fisher Scoring iterations: 4

regTermTest(fit.logit1, test.terms = ~educ, method="Wald", df = NULL)

## Wald test for educ
##  in svyglm(formula = I(deprx == 1) ~ educ, design = des, family = binomial)
## F =  10.18089  on  4  and  19075  df: p= 3.1273e-08

Now, let’s see if, by controlling for other two socio-demographic variables, race and ethnicity and employment status. The fancy word for when an effect is reduced is “attenuated”. We will also do a test to see if the model with the race and ethnicity and employment status significantly improve the model fit. Traditionally, this would be done using a likelihood ratio test, but in survey models, that’s not kosher

#controlling for race/ethnicity and employment status 
summary(fit.logit2)

## 
## Call:
## svyglm(formula = I(deprx == 1) ~ educ + race_eth + empstat, design = des, 
##     family = binomial)
## 
## Survey design:
## svydesign(ids = ~1, strata = ~strata, weights = ~sampweight, 
##     data = sub)
## 
## Coefficients:
##                          Estimate Std. Error t value Pr(>|t|)    
## (Intercept)              -1.54184    0.05926 -26.019  < 2e-16 ***
## educHS Diploma/GED        0.02550    0.06854   0.372    0.710    
## educLess than HS          0.01715    0.09412   0.182    0.855    
## educSome college          0.10796    0.06713   1.608    0.108    
## educUndergraduate Degree -0.07811    0.07158  -1.091    0.275    
## race_ethHispanic         -0.46628    0.10645  -4.380 1.19e-05 ***
## race_ethNH_Black         -0.45330    0.08075  -5.614 2.01e-08 ***
## race_ethNH_Other         -1.11355    0.14608  -7.623 2.60e-14 ***
## empstatUnemployed         0.86237    0.04441  19.417  < 2e-16 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## (Dispersion parameter for binomial family taken to be 0.9961602)
## 
## Number of Fisher Scoring iterations: 4

regTermTest(fit.logit2, test.terms = ~race_eth, method="Wald", df = NULL)

## Wald test for race_eth
##  in svyglm(formula = I(deprx == 1) ~ educ + race_eth + empstat, design = des, 
##     family = binomial)
## F =  33.57233  on  3  and  19071  df: p= < 2.22e-16

After controlling for race/ethnicity and employment status, we see the educational attainment in almost all groups attenuate (reduce in size) somewhat. Hence, the differences in prenatal depression are smaller once controlled for race/ethnicity and employment status. Furthermore, the F test (33.57) suggests that the second model fits the data better than the first one. It is another of these omnibus tests that asks whether there is any variation in our outcome by race and ethnicity and employment status in the second model.

Next we consider the third model, which contains employment and marital status of women:

#controlling for race/ethnicity+employment status+ marital status
summary(fit.logit3)

## 
## Call:
## svyglm(formula = I(deprx == 1) ~ educ + race_eth + +empstat + 
##     mars, design = des, family = binomial)
## 
## Survey design:
## svydesign(ids = ~1, strata = ~strata, weights = ~sampweight, 
##     data = sub)
## 
## Coefficients:
##                          Estimate Std. Error t value Pr(>|t|)    
## (Intercept)              -1.22252    0.07092 -17.237  < 2e-16 ***
## educHS Diploma/GED        0.05230    0.06889   0.759  0.44776    
## educLess than HS          0.02342    0.09534   0.246  0.80595    
## educSome college          0.13679    0.06733   2.031  0.04222 *  
## educUndergraduate Degree -0.05155    0.07188  -0.717  0.47329    
## race_ethHispanic         -0.44134    0.10704  -4.123 3.75e-05 ***
## race_ethNH_Black         -0.42271    0.08159  -5.181 2.23e-07 ***
## race_ethNH_Other         -1.04345    0.14642  -7.127 1.07e-12 ***
## empstatUnemployed         0.80749    0.04605  17.534  < 2e-16 ***
## marsMarried              -0.31549    0.05222  -6.042 1.55e-09 ***
## marsNever Married        -0.64373    0.07065  -9.112  < 2e-16 ***
## marsWidowed              -0.20725    0.07079  -2.928  0.00342 ** 
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## (Dispersion parameter for binomial family taken to be 0.9959921)
## 
## Number of Fisher Scoring iterations: 4

regTermTest(fit.logit3, test.terms = ~empstat, method="Wald", df = NULL)

## Wald test for empstat
##  in svyglm(formula = I(deprx == 1) ~ educ + race_eth + +empstat + 
##     mars, design = des, family = binomial)
## F =  307.4585  on  1  and  19068  df: p= < 2.22e-16

In this model, the effects of employment and marital status of women for educational attainment go back up. This is something confusing. Unemployed women are more likely to take depression medication.

f1 <- fit.logit1   %>%
  tbl_regression(exponentiate = T)

f2 <- fit.logit2   %>%
  tbl_regression(exponentiate = T)

f3 <- fit.logit3   %>%
  tbl_regression(exponentiate = T)

f_all <- tbl_merge( tbls= list(f1, f2, f3),
                    tab_spanner = c("**Model 1**", "**Model 2**", "**Model 3**")) 
f_all

Characteristic	Model 1			Model 2			Model 3
	OR1	95% CI1	p-value	OR1	95% CI1	p-value	OR1	95% CI1	p-value
Education Level
Graduate Degree	—	—		—	—		—	—
HS Diploma/GED	1.21	1.06, 1.38	0.004	1.03	0.90, 1.17	0.7	1.05	0.92, 1.21	0.4
Less than HS	1.25	1.05, 1.49	0.014	1.02	0.85, 1.22	0.9	1.02	0.85, 1.23	0.8
Some college	1.25	1.10, 1.42	<0.001	1.11	0.98, 1.27	0.11	1.15	1.00, 1.31	0.042
Undergraduate Degree	0.91	0.79, 1.04	0.2	0.92	0.80, 1.06	0.3	0.95	0.82, 1.09	0.5
Race/Ethnicity
NH_White				—	—		—	—
Hispanic				0.63	0.51, 0.77	<0.001	0.64	0.52, 0.79	<0.001
NH_Black				0.64	0.54, 0.74	<0.001	0.66	0.56, 0.77	<0.001
NH_Other				0.33	0.25, 0.44	<0.001	0.35	0.26, 0.47	<0.001
Employment Status
Employed				—	—		—	—
Unemployed				2.37	2.17, 2.58	<0.001	2.24	2.05, 2.45	<0.001
Marital Status
Divorced/Separated							—	—
Married							0.73	0.66, 0.81	<0.001
Never Married							0.53	0.46, 0.60	<0.001
Widowed							0.81	0.71, 0.93	0.003
1 OR = Odds Ratio, CI = Confidence Interval

Interpretation:

Model 1 (deprx+educ): Compared to women with graduate level, women with HS diploma/GED (OR =1.21), less than HS (OR = 1.25), and some college degree (OR = 1.25) are statistically more likely to take depression medication. Taking of depression medication among pregnant women with undergraduate degree was not statistically significant (p = 0.2) compared to women with graduate degree, even though results showed that women with undergraduate degree are less likely to take depression medicine.

Model 2: (#educational attainment only +race/ethnicity+employment status): When race/ethnicity and employment is included in the model, along with educational attainment, it was observed that employment status affected likelihood of taking depression medication. Unemployed women are 137% more likely (OR = 2.37) to take depression medication. Furthermore, women from all other racial and ethnic group, NH_Black (0.64), NH_Other(0.33), Hispanic (0.63), are less likely to take depression medication compared to NH_White pregnant women. Interestingly, when race/ethnicity and employment status was included in the model, the impact of educational attainment on taking depression medication was not statistically significant.

Model 3:(educational attainment only +race/ethnicity+employment status+ marital status): Finally, when marital status, race/ethnicity, employment status and educational attainment were included to the model, it was found that compared to divorced women, women who were married (OR = 0.73), never married (OR = 0.53), or widowed (OR = 0.81) all were less likely to take depression medicine. Furthermore, unemployed women (OR = 2.24) were still more likely to use depression medicine and NH_White women were also more likely to take medication for depression compared to women of other race/ethnicity. Interestingly, using this model it was found that statistically, there is 15 % more likelihood of women with some college degree (OR = 1.15) taking the depression medication compared to women with graduate degree.

Comparing models

The Akaike Information Criteria (AIC) test for the models which measures overall model deviance, or residual variance and a penalty term for the number of parameters in a model showed that model three has an the Akaike Information Criteria (AIC) of 19835.9, which is lower than the other two models, indicating model three is the better fit to the data and explains more variation in the data.

AIC(fit.logit1, fit.logit2, fit.logit3)

## Warning in eval(family$initialize): non-integer #successes in a binomial glm!

## Warning in eval(family$initialize): non-integer #successes in a binomial glm!

## Warning in eval(family$initialize): non-integer #successes in a binomial glm!

##          eff.p     AIC deltabar
## [1,]  5.833799 20763.6 1.458450
## [2,] 12.991788 19959.0 1.623974
## [3,] 17.290147 19835.9 1.571832