1. Data summaries

(a) Data Preparation: Choose the interested variables

To investigate the relationship of income gap between men and women and other depending factors. I chose seven representative variables in the NLSY79 dataset(National Longitudinal Survey of Youth, 1979 cohort). Here’s information on what the seven variables mean.

  • gender: (1:male, 2:female)
  • race:(1:other, 2:black, 3:hispanic)
    • income: total income from wages and salary in 2011(top 2% income are topcodded)
    • income_orig: origunal total income from wages and salary in 2011
    • marstat.col_2000: marital status in 2000
      • 1: Never married
      • 2: Married, spouse present
      • 3: other
  • edu_1979: education attainment: HIGHEST GRADE COMPLETED in 2012
    • 1: 1ST GRADE
    • 2: 2ND GRADE
    • 3: 3RD GRADE
    • 4: 4TH GRADE
    • 5: 5TH GRADE
    • 6: 6TH GRADE
    • 7: 7TH GRADE
    • 8: 8TH GRADE
    • 9: 9TH GRADE
    • 10: 10TH GRADE
    • 11: 11TH GRADE
    • 12: 12TH GRADE
    • 13: 1ST YEAR COLLEGE
    • 14: 2ND YEAR COLLEGE
    • 15: 3RD YEAR COLLEGE
    • 16: 4TH YEAR COLLEGE
    • 17: 5TH YEAR COLLEGE
    • 18: 6TH YEAR COLLEGE
    • 19: 7TH YEAR COLLEGE
    • 20: 8TH YEAR COLLEGE OR MORE -nums.jobs: number of jobs worked for in 2000
      -povstatus_1979: family poverty status in 1978(1:in poverty, 0:not in poverty)
## Parsed with column specification:
## cols(
##   .default = col_double()
## )
## See spec(...) for full column specifications.

(b) Transform and relabel gender, race, educational level, crime and marital status variables

## Warning in recode.numeric(.x, !!!values, .default = .default, .missing
## = .missing): NAs introduced by coercion

## Warning in recode.numeric(.x, !!!values, .default = .default, .missing
## = .missing): NAs introduced by coercion
## Warning: Unreplaced values treated as NA as .x is not compatible. Please
## specify replacements exhaustively or supply .default

Note I choose “other” race, male, no crimnal history as our baseline variable(due to prevalance in the dataset)

(c) Compare income gap between men and women

To test whether there is a significant difference of income gap between men and women, first we will plot the income between men and women and compare their income difference.

## Warning: Removed 5662 rows containing non-finite values (stat_boxplot).

We can tell from the boxplot that on average, around 75 percent of male’s annual income is below 75k, whereas 75 percent of female’s annual income is below 50k. Overall, female is associated with lower income. But how can we assess whether this difference is statistically significant? Let’s compute a summary table by adding the standard error(which the standard deviation adjusted by the group size) to assess the statistical significance. 

## # A tibble: 2 x 5
##   gender num.obs mean.income sd.income se.income
##   <fct>    <int>       <dbl>     <dbl>     <dbl>
## 1 Male      6403       53446     69369       867
## 2 Female    6283       29539     35330       446

The summary table suggests that the there’s a relatively small standard error of average female income, which means the sample mean is a more accurate reflection of the actual population mean. The income difference between men and womenis looking quite significant. Now we will run a t test to test our etimates and incorporate the interpretation in our findings

## 
##  Welch Two Sample t-test
## 
## data:  income by gender
## t = 18.034, df = 4993, p-value < 2.2e-16
## alternative hypothesis: true difference in means is not equal to 0
## 95 percent confidence interval:
##  21308.43 26506.37
## sample estimates:
##   mean in group Male mean in group Female 
##             53445.91             29538.51
##  [1] "statistic"   "parameter"   "p.value"     "conf.int"    "estimate"   
##  [6] "null.value"  "stderr"      "alternative" "method"      "data.name"
## [1] 1.752135e-70
##   mean in group Male mean in group Female 
##             53445.91             29538.51
## [1] 21308.43 26506.37
## attr(,"conf.level")
## [1] 0.95

Assessment of Significance - Our study finds that Male on average earn 23907.4dollars higher compared to female annually. (t-statistic 18.03, p=0, 95% CI [21308.4, 26506.4]g).

(d) Test other factors that might influence the income difference

1. race

Here are some questions I will try to answer using the tabular summaries and graphics: - Does race appear to to have an effect on income differnce? - Does income difference appear to be consistent across racial group? - What is the association between race and income difference between men and women?

race gender avg.income
Other Male 69113
Other Female 33570
Black Male 32779
Black Female 24096
Hispanic Male 46017
Hispanic Female 27827

- The tabular table summarize the average income across different groups, and the bargraphs plot the income difference visually. Here are some insights: - Race seems to have an effect on the income difference between men and women. Especially there’s a wide income gap in “other” race, the smallest income difference across racial groups is black, but their average annual income all below $35k. - Income difference appear to be consistent across all racial group: On average male earn higher than female, however, “other” race has a larger income gap across gender: white male almost earn on average double than white female. Hispanic groups and black racial groups also have income difference but not large difference. - From the discussion above we can conclude that race is associated with income difference men and women. ##### 2. marital status Here are some questions I will try to answer using the tabular summaries and graphics: - Does marital status appear to to have an effect on income difference? - Does income difference appear to be consistent across marital status? - What is the association between marital satus and income difference between men and women?

## Warning: Removed 1521 rows containing non-finite values (stat_boxplot).

As we can observe from the group boxplots: - Overall,married couple earn higher than never married people and otehr marital status such as divorce or widowed. - A clear pattern indicates that married male earn more than never married male. However, female income seems to be consistent across all marital status. Married female earn slightly higher than never married and other groups, but not too much income difference. - Marital satus is associated with male’s income, rather than women’s income. ##### 3. Summary table with multiple variables

From the grouping plots we found that the boxplots have a high outlier which we havenot studied well, for those high income earners who earn more than 50k. Wha’s the income performance for men and women from different marital status across racial group?

Here our main variable of interest here is high.income, which indicates whether the individual’s income was over $50K. Anyone for whom high.income == 1 is considered a “high earner”. First, we will create a summary table showing the proportion of high earners varies across all combinations of the following variables: sex, race and marital status.

## # A tibble: 18 x 5
## # Groups:   gender, race [6]
##    gender race     marstat.col_2000        count high.earn.rate
##    <fct>  <fct>    <fct>                   <int>          <dbl>
##  1 Male   Other    Married(spouse present)  1326          0.489
##  2 Male   Other    Never married             296          0.26 
##  3 Male   Other    Other                     366          0.265
##  4 Male   Black    Married(spouse present)   439          0.317
##  5 Male   Black    Never married             433          0.109
##  6 Male   Black    Other                     308          0.182
##  7 Male   Hispanic Married(spouse present)   419          0.394
##  8 Male   Hispanic Never married             160          0.156
##  9 Male   Hispanic Other                     170          0.176
## 10 Female Other    Married(spouse present)  1418          0.198
## 11 Female Other    Never married             172          0.326
## 12 Female Other    Other                     481          0.168
## 13 Female Black    Married(spouse present)   430          0.151
## 14 Female Black    Never married             410          0.12 
## 15 Female Black    Other                     418          0.151
## 16 Female Hispanic Married(spouse present)   447          0.186
## 17 Female Hispanic Never married             107          0.187
## 18 Female Hispanic Other                     230          0.165
gender race marstat.col_2000 count high.earn.rate
Male Other Married(spouse present) 1326 0.489
Male Other Never married 296 0.260
Male Other Other 366 0.265
Male Black Married(spouse present) 439 0.317
Male Black Never married 433 0.109
Male Black Other 308 0.182
Male Hispanic Married(spouse present) 419 0.394
Male Hispanic Never married 160 0.156
Male Hispanic Other 170 0.176
Female Other Married(spouse present) 1418 0.198
Female Other Never married 172 0.326
Female Other Other 481 0.168
Female Black Married(spouse present) 430 0.151
Female Black Never married 410 0.120
Female Black Other 418 0.151
Female Hispanic Married(spouse present) 447 0.186
Female Hispanic Never married 107 0.187
Female Hispanic Other 230 0.165

The summary statistics shows the high earn rate across racial group, marital status, and gender. - White(“Other” race) married male has 48.9 percent of earning more than 50k, followed by married hispanic male, 39.4 percent. Black unmarried male has the lowest earning rate, only 10.9 percent. - Surprisingly, white never married female has the highest earning rate of 32.6 percent, more than white married female of 19.8 percent of high earning rate. Other hgih earn rate difference doesnot vary much across the combination of those variables.

4. Summary bar charts with multiple variables

- The summary bar charts visualize the summary statistics and it reconfirms our conclusion that married male across all racial group tend to have a higher proportion of earning over 50k dollars, however, we don’t observe the difference in female groups except that white never married female has the highest earning rate across all combinations.

5. Test the income gap across different education attainment level

In this part, we will examine the effect of education level on the income gap between men and women by presenting group bars with error bars. We use the respondents’ highest degree obtained in 2011 as the horizontal axis to display the education attainment level, and we caculated the difference of average income between men and women and 95% confidence intervals as our vertical axis to show the height of income gap.

Here are some questions I am interested in:

  • Are there any education degree where women have higher income than men? Are there any degree type appear to not be statistically significantly different between men and women?
  • Which completed education degree appear to have the greatest disparity in income between men and women? What’s the overall significant trend?

Here’s a set of commands that calculates the difference in average income between men and women for each education level

edu_2012 income.gap upper lower is.significant
3rd grade 24742.857 -141360.417 190846.13 0
4th grade -3825.000 -143553.453 135903.45 0
5th grade 18666.667 -23368.389 60701.72 0
6th grade 16918.009 2815.127 31020.89 1
7th grade 4972.159 -4952.595 14896.91 0
8th grade 11912.897 5399.298 18426.50 1
9th grade 14059.536 5231.685 22887.39 1
10th grade 9205.695 3867.041 14544.35 1
11th grade 7851.805 2587.913 13115.70 1
12th grade 14772.958 12473.567 17072.35 1
1st year college 22300.396 14823.478 29777.31 1
2nd year college 21538.275 14685.949 28390.60 1
3rd year college 31804.522 21018.761 42590.28 1
4th year college 52317.325 42056.756 62577.89 1
5th year college 40139.301 22546.934 57731.67 1
6th year college 72665.607 51028.668 94302.55 1
7th year college 54057.965 15521.182 92594.75 1
8th year college 93166.515 61428.526 124904.51 1

  • The tabular summaries and bar charts suggest that among 3rd, 4th, 5th and 7th grade earners, there appears to be not statistically different between men and women in income gaps, which fails to reject the null hypothesis that there’s no income gap between men and women.
  • However, as people completed more advanced education degrees, income gap becomes wider especially during 4th to 8th year+ college, which is between graduate to doctor level studies. Note that 5th year and 7th year college’s error bars seem to overlap with adjacent bars even though there’re showing statistically significance in the test. “8 year plus college” has the greatest parity between men and women, which indicates doctor level female students earn much less than doctor level male students.
6. Test the impact of childhood poverty status on the income gap between men and women

Now we try to test if childhood poverty status mitigate or exacerbate the income gap between men and women, similar apporach to add 95 percent of confidence interval on the error bars of poverty status in 1979.

## Warning: Factor `povstatus_1979` contains implicit NA, consider using
## `forcats::fct_explicit_na`

- This time we found that income gap between men and women appears to be statistically significant for people (not)in poverty because the error bars do not overlap each other, and the income gap for people not in poverty are much higher(almost 30k) than people in poverty in childhood(almost 10k). It indicates that children who were not raised under poverty, men earn much higher than women after they grow up, whereas the poor children don’t have too much income gap between men and women.

2. Mythodologies

(a) Explain various types of analysis

- To investigate the income difference between and women and related factors that influenced the gap, we first compare compare the average income difference between men and women and test its significance leve, then select some interested variables might correlated with income level(gender, income, marital status, race, jobs number, education level, poverty status). For example, we assume that married people will have a wider income difference than unmarried because they some women take traditonal roles of housewives and men are the bread earners. We will compare the marital status across different ethnic groups to test whether the income differnece trend is consistent. - Most of the variables I chose are categorical variables, so I used many bar charts and boxplots to examine the relationship, and to better dignose the income gap, I added the error bars for the education level and poverty status groups. They all proved to be related in the income gap. - Then I ran multi factor regression tests and add interaction to examine the estimates and p-values to assess the significance. - Some dianostic tools are utilized to examine the variance of residuals and clear patterns. Finally we compare the two regression models to test which models suits the dataset the best.

(b) Missing values and the impact

- In the national longitudinal surveys 1979, there are a lot of missingness in the dataset due to the large time span. To better analyze the survey response, fistly, we recode the negative values (refusal, don't know, valid skip, non-interview) as null values, then we divide the variables into two catogories: Factor variables and numeric variables. - For factor variables, we treat missing values as just another factor level, for example. In the marital status, there are around 4700 non-interview respondents, almost one half of the total observations, so we should consdider code them into missing level as jsut another factor leve. Sometimes missingness can be informative or predictive, leading to a significant coeffcient for the missing level. For example, when we ran a logistic regression in which we used missing as one of the marital status to indicate indivisuals whose childhood marital status is unknown, Having poverty satus = missing is likely to be associated with high income gap. - For numeric variables, we simply recode negative values to NA.

(c) Deal with topcoded outcome variables.

- The income variable that we have available is topcoded, which means that for the top 2% of earners, we don’t observe the actual income out of privacy. Instead their income is recoded as the average of the top 2% of incomes. That’s the reason when we always observe outliers which aligned horizontally in many graphs. - Capping is introduced so that model does not learn to correlate extremely high incomes with outcome variable. But at the same time, there could be other variables (say number of children in the households) on which rich people is not going to be an outlier. It is better to keep the record for the person but cap outlier variables like income. To include their data in our regression analysis will help us better converge to the “true” estimates. - So here we compare with the orginal income income.orig and topcoded income income in two regression models used in 3.Findings part(a). 

##                            Estimate Std. Error t value Pr(>|t|)
## (Intercept)               68899.881   1881.970  36.611    0.000
## genderFemale             -24414.487   1455.576 -16.773    0.000
## povstatus_1979in poverty -17175.492   1830.209  -9.384    0.000
## raceBlack                -16488.248   1771.348  -9.308    0.000
## raceHispanic              -7995.504   1972.889  -4.053    0.000
## jobs.num                   -513.028    161.723  -3.172    0.002
##                           Estimate Std. Error t value Pr(>|t|)
## (Intercept)              17509.835   9808.865   1.785    0.081
## genderFemale             -3725.323   5630.802  -0.662    0.512
## povstatus_1979in poverty -4947.737   6685.992  -0.740    0.463
## raceBlack                -1030.302   6543.279  -0.157    0.876
## raceHispanic             -3218.488   8789.524  -0.366    0.716
## jobs.num                   770.167    787.530   0.978    0.334

- The first table is the topcoded income and the second is the orginal values of earned income in 2011. We can tell that original income though has greater values in intercepts, but the estiamtes of coeffcient are much smaller with much bigger standard error, which reflects the true income difference between men and women. However, all the estimates are statitically insignificant(p-values bigger than 0.05). Therefore, topcoded apporach will greatly increase the confidence of interpereting the coefficients and it can generalize to the regression model to test whether there’s is a statitically significant difference in income gap between gender.

(e) Investigated relationships that does not appear in findings sections.

## 
## Call:
## lm(formula = income ~ +gender + edu_2012 + gender * edu_2012, 
##     data = nlsy, na.action = na.exclude)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -165951  -22043   -6055   14598  322349 
## 
## Coefficients:
##                                       Estimate Std. Error t value Pr(>|t|)
## (Intercept)                              34000      34155   0.995 0.319543
## genderFemale                            -24743      38728  -0.639 0.522915
## edu_20124th grade                       -18825      41831  -0.450 0.652705
## edu_20125th grade                       -15333      44094  -0.348 0.728042
## edu_20126th grade                       -12818      37130  -0.345 0.729937
## edu_20127th grade                       -23619      36227  -0.652 0.514440
## edu_20128th grade                       -18052      34729  -0.520 0.603210
## edu_20129th grade                       -12519      34512  -0.363 0.716820
## edu_201210th grade                      -18273      34574  -0.529 0.597151
## edu_201211th grade                      -16021      34472  -0.465 0.642126
## edu_201212th grade                        1594      34176   0.047 0.962805
## edu_20121st year college                 16073      34283   0.469 0.639203
## edu_20122nd year college                 18006      34264   0.526 0.599250
## edu_20123rd year college                 26523      34394   0.771 0.440651
## edu_20124th year college                 65373      34240   1.909 0.056268
## edu_20125th year college                 53995      34579   1.561 0.118455
## edu_20126th year college                 92561      34445   2.687 0.007223
## edu_20127th year college                 90460      34906   2.592 0.009573
## edu_20128th year college                131950      34574   3.816 0.000137
## genderFemale:edu_20124th grade           28568      57006   0.501 0.616289
## genderFemale:edu_20125th grade            6076      58686   0.104 0.917540
## genderFemale:edu_20126th grade            7825      42697   0.183 0.854596
## genderFemale:edu_20127th grade           19771      41854   0.472 0.636671
## genderFemale:edu_20128th grade           12830      40031   0.321 0.748596
## genderFemale:edu_20129th grade           10683      39508   0.270 0.786852
## genderFemale:edu_201210th grade          15537      39528   0.393 0.694282
## genderFemale:edu_201211th grade          16891      39400   0.429 0.668149
## genderFemale:edu_201212th grade           9970      38768   0.257 0.797054
## genderFemale:edu_20121st year college     2442      38925   0.063 0.949969
## genderFemale:edu_20122nd year college     3205      38897   0.082 0.934343
## genderFemale:edu_20123rd year college    -7062      39075  -0.181 0.856593
## genderFemale:edu_20124th year college   -27574      38871  -0.709 0.478115
## genderFemale:edu_20125th year college   -15396      39340  -0.391 0.695540
## genderFemale:edu_20126th year college   -47923      39160  -1.224 0.221082
## genderFemale:edu_20127th year college   -29315      39795  -0.737 0.461360
## genderFemale:edu_20128th year college   -68424      39554  -1.730 0.083698
##                                          
## (Intercept)                              
## genderFemale                             
## edu_20124th grade                        
## edu_20125th grade                        
## edu_20126th grade                        
## edu_20127th grade                        
## edu_20128th grade                        
## edu_20129th grade                        
## edu_201210th grade                       
## edu_201211th grade                       
## edu_201212th grade                       
## edu_20121st year college                 
## edu_20122nd year college                 
## edu_20123rd year college                 
## edu_20124th year college              .  
## edu_20125th year college                 
## edu_20126th year college              ** 
## edu_20127th year college              ** 
## edu_20128th year college              ***
## genderFemale:edu_20124th grade           
## genderFemale:edu_20125th grade           
## genderFemale:edu_20126th grade           
## genderFemale:edu_20127th grade           
## genderFemale:edu_20128th grade           
## genderFemale:edu_20129th grade           
## genderFemale:edu_201210th grade          
## genderFemale:edu_201211th grade          
## genderFemale:edu_201212th grade          
## genderFemale:edu_20121st year college    
## genderFemale:edu_20122nd year college    
## genderFemale:edu_20123rd year college    
## genderFemale:edu_20124th year college    
## genderFemale:edu_20125th year college    
## genderFemale:edu_20126th year college    
## genderFemale:edu_20127th year college    
## genderFemale:edu_20128th year college .  
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 48300 on 6986 degrees of freedom
##   (5664 observations deleted due to missingness)
## Multiple R-squared:  0.2558, Adjusted R-squared:  0.2521 
## F-statistic: 68.63 on 35 and 6986 DF,  p-value: < 2.2e-16

- We also tried to involve other factor variables in our regression models, such as interaction between education level and gender. We ran the same regression model to examine the p-values of genderFemale:education to be not sinificant, which means education attainemnt does positively relates to income level, but it doesn’t relate to the income difference between men and women. Education can affect income, but it does not affect the income fact, so it cannot exacerbate or mitigate income gap between men and women. This is the difference between significant main effects and significant interactions.

3. Findings

(a) Assessing significance of factors in linearility regression

Assessment of Significance of income difference between men and women - Our study finds that Male on average earn 23907.4dollars higher compared to female annually. (t-statistic 18.03, p=0, 95% CI [21308.4, 26506.4]g).

To examine which variable is likely to be statistically significant predictor of income difference between men and women, here we include poverty status, race and gender and jobs.num in our linear regression model.

We will try to answer the following questions: - What is the interpretation of the coefficient of jobs.num in this model? - What is the interpretation of the coefficient of poverty status?

Estimate Std. Error t value Pr(>|t|)
(Intercept) 68899.88 1881.97 36.61 0.00000
genderFemale -24414.49 1455.58 -16.77 0.00000
povstatus_1979in poverty -17175.49 1830.21 -9.38 0.00000
raceBlack -16488.25 1771.35 -9.31 0.00000
raceHispanic -7995.50 1972.89 -4.05 0.00005
jobs.num -513.03 161.72 -3.17 0.00152

- Because there are three categorical variables in our regression model, we need to specify the baseline before giving interpretation of other three coefficients.The baseline is white male not in poverty during childhood with zero job, so the estimate of raceblack means that the estimated intercept is 16488 dollars higher among white compared to black. Similarly, the estimated of genderFemale means that the estimated intecept is 17175 dollars higher among non-poverty earners than poverty earners. - Another way of putting it: For two people of the same race and gender and same poverty status, every additional prior job is on average associated with a $513 decrease in income. - Among people of the same race and gender who have previously held the same number of jobs, people living in poverty on average $17175 less than people who are not raised in poverty. - Looking at the p-values, it looks like gender, povstatus_1979 (childhood poverty status in 1979), race and job numbers are all statistically significant predictors of indivisual income. #### (b) Diagnostic tools to assess whether the linear model is apporpirate.

- The first two plots are the most important, but the last two can also help with identifying outliers and non-linearities. - Residual vs. Fitted Plot: There is a clear non-linearity present in the plot. we see that the variance appear to be increasing in fitted value in a horizontal funnel shape and there are plenty of outlier residuals. - Normal QQ plot: the underlying normality assumptions don’t hold here, the residuals appear highly non-normal. Both the lower tail and upper tail are heavier than we would expect under normality and we found an isolated upper tail, is correlated the outlier in the residual and fitted plot. - Scale and Location plot: There is a slight indication of non-constant (heteroskedastic) variance. - Residuals vs Leverage: There appear to be clear outliers in the data.

(c) Update linear regression model from part (a) to also include an interaction term between race and gender and compare the two models

Note: The interactive model will better estimate the income gap between men and women rather just evaluate the average income difference.
Estimate Std. Error t value Pr(>|t|)
(Intercept) 73514.25 1973.08 37.26 0.00000
genderFemale -34925.51 2032.41 -17.18 0.00000
povstatus_1979in poverty -17844.20 1822.62 -9.79 0.00000
raceBlack -29428.94 2429.40 -12.11 0.00000
raceHispanic -15107.16 2760.96 -5.47 0.00000
jobs.num -416.85 161.38 -2.58 0.00982
genderFemale:raceBlack 25739.18 3323.67 7.74 0.00000
genderFemale:raceHispanic 14444.32 3823.94 3.78 0.00016 
## Analysis of Variance Table
## 
## Model 1: income ~ +gender + povstatus_1979 + race + jobs.num
## Model 2: income ~ gender + povstatus_1979 + race + jobs.num + gender:race
##   Res.Df            RSS Df    Sum of Sq      F    Pr(>F)    
## 1   5461 15565922814681                                     
## 2   5459 15392073630859  2 173849183822 30.829 4.855e-14 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

- In the interaction term, there are two estimated coefficients added in the model.For example, the coeffcient of genderFemale:raceBlack indicates that among people of the same poverty status and same number of jobs, the average income gap between men and women is 25738 dollars lower in black racial groups than in the other racial group. - Looking at the p-values, even for adjusting for poverty status and number of jobs, there is a statistically significant difference in the income gap between men and women across racial groups. Income gap among other racial group is much wider than black and hispanic groups. - The interaction between race and gender turns out to be a highly statistically predictor of income gap in the model. We can see smaller standard errors in the interactive models. One we control for poverty status and number of prior jobs, our data is consistent with the income gap between men and women being the same across different racial groups in the U.S.

4. Discussions

- I have around 70 percent of confidence in my analysis and they’re statistically significant in my opinion. I believe that race, childhood poverty status and marital status are related to income gap between men and women because multiple tests and graphs with error bars proved my view point. I hope those findings can help establish more women empowerment policies to benefit married women and people who were raised up in poverty of different race.