Bivariate Analyses Chisquare and Cramers V

PRAMS8 Maternal Health Dataset - Bivariate Analyses

We conducted bivariate analyses using the chi-square hypothesis test to determine which variables in the PRAMS dataset were statistically associated with maternal health depression. Variables were selected based on hypothesized relationships laid out in the structural equation model framework developed by the PI. Variable selection was also informed by updated findings from the univariate analyses.

Variables Selected for Bivariate Analysis

Direct Effect Model Components and Variables to Model Social Inequality

  • Socio-Economic Status/Education: Education (MAT_ED), Income (income.label), MARRIED

  • Age (age.label)

  • Mediator Variable - Race/Ethnicity (MRACE_AMI, MRACE_BLK, MRACE_WHT, MRACE_ASIAN_PU, HISP_BC, MRACE_NHOPI_PU, MRACE_MULTO_PU)

  • Outcome Variable - Post-Partum Maternal Depression (MM_PPDR2)

Indirect Effect Model Components and Variables to Model Social Context

  • Variables above including

  • Discrimination, which was replaced with History of Interpersonal Violence variables: IPV_Husband, IPV_Exhusband, IPV_Family Member, and IPV_NonFamilyMember after univariate analyses (e.g., one-way frequency count tables and bar plots) found no viable observations to support inclusion of a Discrimination variable.

  • Health Insurance Status: (Employer, Parental, Health Exchange, Medicaid, Military/TRICARE, IHS/Tribal Services, Other, Uninsured, CHIP, Other Government Assistance, which will be transformed for modeling as one health insurance variable with four levels: Private, Public, Other, or Uninsured.

  • Social Support, which may be replaced with Resource Environment to conceptualize mothers who receive WIC (MAT_WIC) as a model variable.

  • Pregnancy Intent (PGINTENT), which was added to the model after univariate analysis and literature review which supported adding this variable into our model to understand the relationship between a mother’s intent to become pregnant and post-partum maternal depression.

  • History of Pre-Pregnancy Depression and symptoms of post-partum mental health depression (BGP_DEPRS82, MH_PPINT2)

  • Substance Use expressed as history of smoking cigarettes (SMK2YRS2, SMK63B_A2) and history of and frequency of alcohol use (DRK_2YRS2 and DRK8_3B2).

Key Takeaways - Relationships of Selected Variables to Post-Partum Maternal Depression

  • Pregnancy Intentions, reported PPD symptoms, history of depression before pregnancy, and income have both a statistically significant relationship with post-partum maternal depression and medium-to-strong association with post-partum maternal depression compared to all other variables as indicated by test statistics, p-values, and calculated effect sizes.

  • Pregnancy intention has the strongest association with post-partum maternal depression in this dataset.

  • Other variables with statistically significant relationships and detectable effect sizes include marital status and substance use (i.e., smoking in the past 2 years, smoking at all pre-pregnancy, and drinking in the past 2 years).

  • Age, drinking frequency, and history of interpersonal violence had negligible effects but statistically significant relationships.

  • Race and insurance did not have a statistically significant relationship or observable effect on post-partum maternal depression.

Interpreted Bivariate Analysis Results
Variable

Chi-Square

p-value

Statistically Significant? Cramer’s V

Effect Size Interpretation

DF = 2

Race/Ethnicity

X2 = 0

p = 1

No 0 No association with PPD
Marital Status

X2 = 1677.5,

p = >0.005

Yes 0.09 Small association with PPD
Age

X2 = 233.33

p = >0.005

Yes 0.03 Negligible association with PPD
Income

X2 = 21124

p = >0.005

Yes 0.22 Medium association with PPD
Insurance

X2 = 0

p = 1

No 0 No association with PPD
Smoking History - Past 2 Years

X2 = 2898.5

p = >0.005

Yes 0.12 Small association with PPD
Smoking History Pre-Pregnancy at All

X2 = 2677.07

p = >0.005

Yes 0.11 Small association with PPD
Drinking History - Past 2 Years

X2 = 1178.5

p = >0.005

Yes 0.07 Small association with PPD
Drinking History - Frequency

X2 = 1457.4

p = >0.005

Yes 0.06 Small association with PPD
History of Interpersonal Violence

X2 = 32.057

p = >0.005

Yes 0.004 Negligible association with PPD
Intent to Become Pregnant

X2 = 190111

p = >0.005

Yes 0.66 Largest association with PPD
History of Depression Before Pregnancy

X2 = 20334

p = 0.005

Yes 0.30 Medium association with PPD
Reports PPD Symptoms

X2 = 53548

p = 0.005

Yes 0.49 Large association with PPD

Hypothesis Testing with Chi-Square and Effect Size Calculation with Cramer’s V

The chi-square test was used to confirm the hypothesized relationship between our predictor variables and post-partum maternal health depression. The null hypothesis asserts that there is no relationship between our variables and maternal health depression, while the alternative hypothesis supports a rejection of the null in favor of evidence that there is a relationship between our variables and maternal health depression.

Cramer’s V was used to measure the effect size between variables we hypothesize to be associated with maternal post-partum depression. The larger the effect size, the stronger the association between two variables. While the hypothesis test can be used to test which variables are related in our dataset, the effect size can also tell us how big or small the association is between those variables. Further, some survey items in the PRAM dataset may have less responses than others, and since sample sizes can change between two variables due to variations in responses across respondents, then the effect size can be a more robust measure of the hypothesized relationship between two variables compared to the p-value of statistical significance, which is less likely to detect stastically significant relationships for smaller samples.

#Environmental Setup and Packages 
install.packages("pacman", repos='http://cran.us.r-project.org')
Installing package into 'C:/Users/jaymo/AppData/Local/R/win-library/4.5'
(as 'lib' is unspecified)
package 'pacman' successfully unpacked and MD5 sums checked

The downloaded binary packages are in
    C:\Users\jaymo\AppData\Local\Temp\RtmpG0Z3Am\downloaded_packages
library(pacman)
Warning: package 'pacman' was built under R version 4.5.2
p_load("tidyverse", "ca", "factoextra", "lsr")
setwd("C:/Users/jaymo/OneDrive/SUPERNOVA/Output Files/Datasets")
prams.transformed <- read.csv("TransformedPRAMS8Variables.csv") 

Race and Depression - Accept the Null Hypothesis and No Association Detected

prams.race.long <-  prams.transformed %>% 
  select(MRACE_AMI:HISP_BC, MH_PPDPR2) %>%
  pivot_longer(!MH_PPDPR2,
               names_to = "Race/Ethnicity",
               values_to = "Response")


race.contingency.table <- table(prams.race.long$`Race/Ethnicity`, prams.race.long$MH_PPDPR2)

chisq.test(race.contingency.table)

    Pearson's Chi-squared test

data:  race.contingency.table
X-squared = 0, df = 12, p-value = 1
cramersV(race.contingency.table)
[1] 0

Marital Status and Depression - Reject the Null Hypothesis and Small Association Detected

prams.marital <- prams.transformed %>% select(MARRIED, MH_PPDPR2) %>%
  filter(MARRIED != "UNKNOWN")

chisq.test(table(prams.marital$MARRIED, prams.marital$MH_PPDPR2))

    Pearson's Chi-squared test

data:  table(prams.marital$MARRIED, prams.marital$MH_PPDPR2)
X-squared = 1677.5, df = 2, p-value < 2.2e-16
cramersV(table(prams.marital$MARRIED, prams.marital$MH_PPDPR2))
[1] 0.08708612

Age and Depression - Reject the Null Hypothesis and Negligible Association Detected

prams.age <- prams.transformed %>% select(age.label, MH_PPDPR2) %>%
  filter(age.label != "Unknown")

chisq.test(table(prams.age$age.label, prams.age$MH_PPDPR2))

    Pearson's Chi-squared test

data:  table(prams.age$age.label, prams.age$MH_PPDPR2)
X-squared = 233.33, df = 2, p-value < 2.2e-16
cramersV(table(prams.age$age.label, prams.age$MH_PPDPR2))
[1] 0.03326785

Income and Depression - Reject the Null Hypothesis and Medium Association Detected

prams.income <- prams.transformed %>% select(income.label, MH_PPDPR2)

chisq.test(table(prams.income$income.label, prams.income$MH_PPDPR2))

    Pearson's Chi-squared test

data:  table(prams.income$income.label, prams.income$MH_PPDPR2)
X-squared = 21124, df = 6, p-value < 2.2e-16
cramersV(table(prams.income$income.label, prams.income$MH_PPDPR2))
[1] 0.2184242

Insurance and Depression - Accept the Null and No Association Detected

prams.insurance <- prams.transformed %>% select(MH_PPDPR2 ,Employer:GOV2)  

prams.insurance <- prams.insurance %>% pivot_longer(!MH_PPDPR2,
                                        names_to = "Insurance",
                                        values_to = "Response") 

prams.insurance.transformed <- prams.insurance %>% mutate(insurancetype = 
                                                recode(Insurance,
                                                       "Medicaid" = "Public Insurance Coverage",
                                                       "IHS" = "Public Insurance Coverage",
                                                       "Employer" = "Private Insurance Coverage",
                                                       "Parent" = "Other Insurance Coverage",
                                                       "Military_TRICARE" = "Public Insurance Coverage",
                                                       "Uninsured" = "Uninsured",
                                                       "Other" = "Other Insurance Coverage",
                                                       "HealthExchange" = "Private Insurance Coverage",
                                                       "CHIP" = "Public Insurance Coverage",
                                                       "GOV1" = "Public Insurance Coverage",
                                                       "GOV2" = "Public Insurance Coverage")) 

table(prams.insurance.transformed$MH_PPDPR2, prams.insurance.transformed$insurancetype)
                       
                        Other Insurance Coverage Private Insurance Coverage
  BLANK/DK                                  9690                       9690
  Does Not Indicate PPD                   300922                     300922
  Indicates PPD                           132150                     132150
                       
                        Public Insurance Coverage Uninsured
  BLANK/DK                                  29070      4845
  Does Not Indicate PPD                    902766    150461
  Indicates PPD                            396450     66075
chisq.test(table(prams.insurance.transformed$MH_PPDPR2, prams.insurance.transformed$insurancetype))

    Pearson's Chi-squared test

data:  table(prams.insurance.transformed$MH_PPDPR2, prams.insurance.transformed$insurancetype)
X-squared = 0, df = 6, p-value = 1
cramersV(table(prams.insurance.transformed$MH_PPDPR2, prams.insurance.transformed$insurancetype))
[1] 0

Substance Use - Smoking History in Past 2 Years - Reject the Null and Small Association Detected

prams.smoke2years <- prams.transformed %>%
  select(MH_PPDPR2, SMK2YRS2) %>%
  filter(SMK2YRS2 != "DK/BLANK")
  
chisq.test(table(prams.smoke2years$MH_PPDPR2, prams.smoke2years$SMK2YRS2))

    Pearson's Chi-squared test

data:  table(prams.smoke2years$MH_PPDPR2, prams.smoke2years$SMK2YRS2)
X-squared = 2898.5, df = 2, p-value < 2.2e-16
cramersV(table(prams.smoke2years$MH_PPDPR2, prams.smoke2years$SMK2YRS2))
[1] 0.1151662

Substance Use - Smoking History Pre-Pregnancy - Reject the Null and Small Association Detected

prams.smok.prepreg <- prams.transformed %>%
  select(MH_PPDPR2, SMK63B_A2) %>%
  filter(SMK63B_A2 != "DK/BLANK")

chisq.test(table(prams.smok.prepreg$MH_PPDPR2, prams.smok.prepreg$SMK63B_A2))

    Pearson's Chi-squared test

data:  table(prams.smok.prepreg$MH_PPDPR2, prams.smok.prepreg$SMK63B_A2)
X-squared = 2677.9, df = 2, p-value < 2.2e-16
cramersV(table(prams.smok.prepreg$MH_PPDPR2, prams.smok.prepreg$SMK63B_A2))
[1] 0.11078

Substance Use - Drinking History Past 2 Years - Reject the Null and Small Association Detected

prams.drink2years <- prams.transformed %>% 
  select(MH_PPDPR2, DRK_2YRS2) %>%
  filter(DRK_2YRS2 != "DK/BLANK")

chisq.test(table(prams.drink2years$MH_PPDPR2, prams.drink2years$DRK_2YRS2))

    Pearson's Chi-squared test

data:  table(prams.drink2years$MH_PPDPR2, prams.drink2years$DRK_2YRS2)
X-squared = 1178.5, df = 2, p-value < 2.2e-16
cramersV(table(prams.drink2years$MH_PPDPR2, prams.drink2years$DRK_2YRS2))
[1] 0.07350205

Substance Use - Drinking Frequency - Reject the Null and Negligible Association Detected

prams.drink.freq <- prams.transformed %>%
  select(MH_PPDPR2, DRK8_3B2) %>%
  filter(DRK8_3B2 != "DK/BLANK")

chisq.test(table(prams.drink.freq$MH_PPDPR2, prams.drink.freq$DRK8_3B2))

    Pearson's Chi-squared test

data:  table(prams.drink.freq$MH_PPDPR2, prams.drink.freq$DRK8_3B2)
X-squared = 1457.4, df = 10, p-value < 2.2e-16
cramersV(table(prams.drink.freq$MH_PPDPR2, prams.drink.freq$DRK8_3B2))
[1] 0.05786952

History of Interpersonal Violence - Reject the Null and Negligible Association Detected

prams.ipv.long <- prams.transformed %>%
  select(MH_PPDPR2, IPV_Husband:IPV_NonFamilyMember) %>%
  pivot_longer(!MH_PPDPR2, 
               names_to = "IPV History",
               values_to = "Response") %>%
  filter(Response != "BLANK/DK")

chisq.test(table(prams.ipv.long$MH_PPDPR2, prams.ipv.long$`IPV History`))

    Pearson's Chi-squared test

data:  table(prams.ipv.long$MH_PPDPR2, prams.ipv.long$`IPV History`)
X-squared = 32.057, df = 6, p-value = 1.591e-05
cramersV(table(prams.ipv.long$MH_PPDPR2, prams.ipv.long$`IPV History`))
[1] 0.004802427

Intent to become Pregnant - Reject the Null and LARGEST Association Detected

prams.preg.intentions <- prams.transformed %>%
  select(PGINTENT2, MH_PPDPR2) %>%
  filter(PGINTENT2 != "BLANK")

chisq.test(table(prams.preg.intentions$PGINTENT2))

    Chi-squared test for given probabilities

data:  table(prams.preg.intentions$PGINTENT2)
X-squared = 190111, df = 2, p-value < 2.2e-16
cramersV(table(prams.preg.intentions$PGINTENT2))
[1] 0.6609124

History of Depression Before Pregnancy - Reject the Null and Medium Association Detected

prams.prepreg.dep <- prams.transformed %>% select(MH_PPDPR2, BPG_DEPRS82) %>%
  filter(BPG_DEPRS82 != "DK/BLANK")

chisq.test(table(prams.prepreg.dep$MH_PPDPR2, prams.prepreg.dep$BPG_DEPRS82))

    Pearson's Chi-squared test

data:  table(prams.prepreg.dep$MH_PPDPR2, prams.prepreg.dep$BPG_DEPRS82)
X-squared = 20334, df = 2, p-value < 2.2e-16
cramersV(table(prams.prepreg.dep$MH_PPDPR2, prams.prepreg.dep$BPG_DEPRS82))
[1] 0.3045234

PPD Symptoms - Reject the Null and LARGE Association Detected

prams.symptoms <- prams.transformed %>%
  select(MH_PPDPR2, MH_PPINT2) %>%
  filter(MH_PPINT2 != "BLANK/DK")

chisq.test(table(prams.symptoms$MH_PPINT2, prams.symptoms$MH_PPDPR2))

    Pearson's Chi-squared test

data:  table(prams.symptoms$MH_PPINT2, prams.symptoms$MH_PPDPR2)
X-squared = 52538, df = 2, p-value < 2.2e-16
cramersV(table(prams.symptoms$MH_PPINT2, prams.symptoms$MH_PPDPR2))
[1] 0.4924697

Conclusion