finalprojecteconc142
TRANSCRIPT
Luke Chen
Econ C142 Final Project
Table 1:
All female male
Ttest (female vs male)
mean se mean se mean se t pvalue
Educ8.7239
9 0.02472 9.1515 0.04287
8.49442
0.03006 12.5489 < 2.2e-16
Age36.404
7 0.05502 36.3151
0.09216
36.4528
0.06857
-29.8542 < 2.2e-16
Y1.5671
6 0.00327 1.43867
0.00521
1.63616
0.00407 -1.1991 0.2305
Va3.0715
4 0.0038 2.94475
0.00677
3.13963
0.00447
-24.0223 < 2.2e-16
Education female male Y female male
418.13
0119.64
4 quantile 1
39.6859
17.0939
621.01
1626.66
36 quantile 2
22.2731
26.4843
920.73
9524.11
34 quantile 3
19.7502
27.8125
1227.99
919.84
99 quantile 4
18.2909
28.6094
1612.11
979.729
05 total 100 100 100 100
The blue smoothed histogram looks somewhat like the red one shifted to the left.
It is clear from the smoothed histograms that the distribution for the blues, while having a range very like that of the distribution for the reds (around 1.6 to 4.5), is much more skewed to the right, while the distribution of the reds is much more skewed to the left. This is one way to show va for females tends to be lower than va for males.
Also, a more obvious statistic is : both the female wage and va means are lower than the male wage and va means.
Table 2:
Coefficients: constant female educ age^3 s1 1.636162 -0.197495 s2
0.7791 -0.2496 0.082720.00000273
2 s3
0.5115 0.085520.00000259
6 s4
0.7893 0.081020.00000280
7
a. Both coefficients are negative in S1 and S2. After education and age effects are considered in S2, the coefficient is more negative in the S2 model
b. Coefficients for education and age are similar between female and male, while those for male are slightly larger. Oaxaca decomposition is done as below. When using female beta, the mean difference contributes 0.056 (education), 0.002(age), beta difference contributes 0.04(education), 0.01(age). Sum of them and the difference between constants is equal to 0.197, which explains all the wage difference between male and female.
Mean and Coefficients:
wage educ_mean age^3_mean educ_coef age^3_coef c female 1.438668 9.151496 55695.5 0.08552 0.000002596 0.5115 male 1.636162 8.494422 56518.37 0.08102 0.000002807 0.7893 diff 0.197 -0.657074 822.87 -0.0045 0.000000211 0.2778
Oaxaca Decomposition:
educ age const
mean diff (x_b-x_a)*beta_a
coef diff x_b*(beta_b-beta_a) mean diff coef diff const diff total
deco female beta
-0.056192968 -0.038224899 0.002136171 0.011925376 0.2778 0.197
deco male beta
-0.053236135 -0.041181732 0.002309796 0.011751751 0.2778 0.197
The contribution of education is larger in absolute value than the contribution of the coefficient on it, but in an opposite sign of the total difference for both the female beta and male beta decompositions. The contribution of age is in the same direction (positive) as the contribution of the coefficient on it, and both are in the same direction as the final difference, but the absolute value of the age contribution is smaller than the absolute value of the contribution of the coefficient on it. This shows how education seems to contribute oppositely to the gender gap, while age difference contributes positively but the returns to age difference contribute more than the age difference itself.
Figure 2: Women and Men Plots for each of the 5 Education Levels (with regression lines)
Figure 2b) Plots of each of the 5 Education Levels for both Men and Women(blue for men, red for women)
For both male and female, the best choice of K is K=2.
Figure 3 - Mean Squared Errors for Different K Values
Figure 4A and B : Spline v Cubic Regressions
Table 3:
Coeffs:
c female educ age^3 va M1
0.77910 -0.24960 0.08272 0.0000027
M2
0.13990 -0.19550 0.07060 0.0000027 0.23690
M3 (male)
0.50660 0.07049 0.0000028 0.25880
M4(female)
0.39640 0.07151 0.0000025 0.20080
M1:
Y = -0.25 * female + 0.08 * educ + 0.0000027 * age^3 + 0.779
M2:
Y = -0.19 * female + 0.07 * educ + 0.00000027 * age^3 +0.24*va+ 0.139
Gender gap explained by women working in less productive employer:
0.34938 * (-0.19 + 0.25) = 0.0189Where 0.34938 is the mean of femaleTotal gender gap is 1.636-1.439=0.197
About 0.0189/0.197 = 9.6% is explained by VA
Define: ~β as the estimate from (1), β as the estimate from OLS. Since a (productivity factor) is positively related to wage y, i.e. a > 0, we have ~β>β
Therefore if more productive employers hire workers with higher productivity, the workers’ wages tend to be higher.
Mean and Coefficients:
log wageeduc_mean
age^3_mean va_mean
educ_coef age^3_coef va_coef c
female
1.43866
8 9.151496 55695.5 2.944748 0.07151 2.49E-06 0.2008 0.05431 male
1.63616
2 8.494422 56518.37 3.139626 0.07049 2.82E-06 0.2588 0.06563 diff 0.197 -0.65707 822.87 0.194878 -0.00102 3.26E-07 0.058 0.01132
Oaxaca Decomposition:
educ age va const
mean diff (x_b-x_a)*beta_a
coef diff x_b*(beta_b-beta_a)
mean diff coef diff
mean diff coef diff
const diff total
deco female beta -0.04699 -0.00866
0.002048
0.018425
0.039132
0.182098
0.01132
0.197
deco male beta -0.04632 -0.00933
0.002316
0.018157
0.050434
0.170795
0.01132
0.197
After renormalization (subtracting the lowest value of VA)
Mean and Coefficients:
log wage educ_mean age^3_mean va_mean educ_coef age^3_coef va_coef c female 1.438668 9.151496 55695.5 1.241117 0.07151000 0.00000249 0.20080000 0.39640000 male 1.636162 8.494422 56518.37 1.435995 0.07049000 0.00000282 0.25880000 0.50660000 diff 0.197 -0.657074 822.87 0.194878 -0.00102 3.26E-07 0.058 0.1102
Oaxaca Decomposition:
educ age va const
mean diff (x_b-x_a)*beta_a
coef diff x_b*(beta_b-beta_a) mean diff
coef diff
mean diff coef diff
const diff total
deco female beta -0.04699 -0.00866 0.002048
0.018425
0.039132
0.083288 0.1102 0.197
deco male beta -0.04632 -0.00933 0.002316
0.018157
0.050434
0.071985 0.1102 0.197
Table 4:
Coefficients:
c age age^2 female dva C1
0.258371
2 -0.0099603 0.0001025 0.0759453
C20.422600
0 0.0635200 -0.0007875 -0.1981000 0.0034580
C3 (male)0.276200
0 -0.0102200 0.0000993 0.0831200
C4(female)0.227200
0 -0.0096330 0.0001108 0.0621000
The coefficients on dva are indeed all positive.
The coefficients for dva are smaller (<0.1) compared with those for va in M2-M4 (>0.2).
The reason is that age is now negatively correlated with dy.
In both models M3 and M4, and models C3 and C4, the coefficients of va for women are smaller than those for men. In other words, even though women work in more productive firms, they are not likely to be paid the same as men. Women may work in less-paid jobs in the same firm, or receive lower salary for the same jobs.
Table 5:
va_previous va diff
women men women menwomen men
Q10.31388 0.2157 0.3469 0.1979
70.0330
2-
0.01773
Q20.2252 0.2633
20.2159
30.2447
2-
0.00928
-0.01859
Q30.24796 0.2507
60.2355
90.2813
1-
0.01237
0.030549
Q40.21296 0.2702
20.2015
8 0.276-
0.01138
0.005778
total 1 1 1 1 0 0
female male from to yl2 yl1 y yp1 yl2 yl1 y yp1
1 1 1.090181.09661
41.12507
71.14801
61.23042
51.23230
91.26972
31.29043
5
1 21.20107
51.20394
81.26198
71.29403
81.33349
41.32995
31.40003
21.42845
1
1 31.24805
11.25164
41.33248
61.35729
21.36298
21.35689
81.48636
81.52182
8
1 41.40351
71.40779
61.50730
91.54856
71.46838
11.46010
1 1.6251.65862
3
2 11.23535
81.24161
31.22686
91.24434
91.35510
11.35717
31.33932
71.36196
5
2 21.28071
91.30735
71.33806
11.36992
91.42894
11.43285
8 1.477141.49397
3
2 31.38250
1 1.37693 1.438081.45534
41.48321
51.49157
61.56388
61.59612
9
2 41.39452
51.40475
31.49381
11.52014
51.52464
6 1.529761.64534
81.67828
7
3 11.31651
51.34062
4 1.275211.30323
11.50454
61.50885
81.45681
71.46240
8
3 21.44741
4 1.448481.48400
11.53020
11.55141
11.55574
21.58218
21.61434
7
3 31.52172
91.50726
61.56762
11.55747
91.64572
21.64623
51.69957
31.71329
3
3 41.64459
71.66506
41.76400
71.78999
41.69404
11.70706
31.80547
21.84444
2
4 11.47464
81.49296
41.43655
61.45238
7 1.664671.67769
61.58318
61.61032
6
4 21.53662
7 1.545731.50803
71.54046
31.70081
21.71045
41.67279
51.69709
5
4 31.69258
41.71482
81.72946
71.75690
81.79522
51.80200
61.81722
51.84842
2
4 41.94395
11.96652
62.02591
92.05965
62.04960
12.06483
42.14593
12.17629
5
If employer productivity has a positive effect, the wage change from lower quantile to higher quantile should be positive, and negative in the other way. Check the change from quantile 1 to 4 for female, it is indeed positive (1.507309-407796), and the change from quantile 4 to 1 it is negative (1.436556 - 1.492964). The pattern for men is similar.
My analysis: If employer productivity has a positive causal effect on wages, the profiles for wage change should reflect a positive trend for workers who move to higher quantiles between different time intervals, and a negative trend for workers who move to lower quantiles between different time periods. We can check this by examining the mean differences between y and yl1 for females and males for those who moved from quantile 1 to 4 and 4 to 1 respectively. For females, this gives (1.507309-.407796) and (1.436556 - 1.492964). These indeed are positive and negative values.
Figure 5
Figure 6
By looking at Table 5 and Figure 5 and 6, we notice Moving from lower quantile to higher, wage increases. True for both men and women.
Wage changes are bigger for men. Change from 4->1 is different from 1->4, i.e. not symmetric. More men than women move from lower quantiles to the higher ones
We conclude that employer productivity is a factor for employee wage and male workers benefits more from moving from lower productivity employers to the higher ones.
My words: The data to tend to support that moving to higher productivity firm increases wages while moving to a lower productivity firm decreases wages, as evidenced by differences being positive and negative respectively for the two scenarios. This is true for both women and men. The figures also show the positive part of this relationship due to positive slopes (although not perfectly linear) for men and women who went through all the quantile increases (destination quantile 2, 3, 4 but not 1). However, the increase in wages for men due to moving to a higher productivity employer is higher than that for women, as suggested by comparing the slopes in the corresponding plots in Figures 5 and 6 (Figure 6 has smaller slopes), suggesting that women benefit less no matter what destination quantile from a quantile increase in employer productivity.
Table 6:
Mean and Coefficients:
Wage wage (no va) educ_mean age^3_mean educ_coef age^3_coef c
female
1.438668 1.255799 9.151496 55695.5 0.08118 0.000002563 0.3701
male
1.636162 1.375197 8.494422 56518.37 0.07764 0.00000281 0.5569
diff 0.197494 0.119 -0.657074 822.87 -0.00354 0.000000247 0.1868
Decomposition
educ age const
mean diff (x_b-x_a)*beta_a
coef diff x_b*(beta_b-beta_a) mean diff coef diff const diff total
deco female beta -0.05334
-0.030070254 0.00210902 0.013960037 0.1868 0.119
deco male beta -0.05102
-0.032396296 0.00231226 0.013756789 0.1868 0.119
VA contributions
va mean va coef va contribution
female 2.94474 0.0621000 0.182868851
8 0
male3.13962
60.0831200
0 0.260965713
male - female 0.078096862
Note 0.078096862 (gender gap of va) + 0.119457532 (gender gap without va) = 0.197494 (gender gap)
0.078096862/0. 197494 = 40%
Therefore about 40% of the gender gap is due to VA.
To decompose this into “due to different va means for male and female”, and “due to different returns to va for male and female”, we could hold the mean constant at the average of the 2 means, and then the coeff constant at the average of the 2 coeffs, and find the respective va contributions. Then difference them out, and divide those differences by the gender gap, which is .197494 to get the two respective percentages (first the coeff on va difference contribution and second the difference in mean va contribution).
Conclusion :
There is evidence that men and women are paid differently. Productivity of employer is a factor for a worker’s wage. Moving from lower productivity employers to higher productivity employers results in a wage
increase. Women benefit less from moving from lower productivity employers to the higher ones. This is
likely due to pay difference in the same employers. Less women move from less productive employers to the more productive ones. Productivity of employer is positively correlated to individual’s productivity.