Jeremiah Rice

Jeremiah Rice

December 2, 2013

Applied Econometrics Fall 2013

Married Women in the Labor Force Analysis

In my research paper we will be looking specifically at married women in the labor force. This is a very interesting topic since women and their wages have always been compared to men, but the details of a woman’s wage, and why it might be different, never really defined.Of course an employer might give multiple reasons if you ask them why a woman makes less than her male counterpart, and whether those reasons are justified or not is a completely separate argument in itself. What I do want to focus on is married women in the labor force who have children, and what there wage, or incomes, are. I believe that maternity leave is a legitimate reason why a woman might get paid less, but I believe that having children affects all women in a much more drastic way then you might think. For example, it might keep a woman out of the labor force all together.

My hypothesis is that, women who are highly paid in the labor force choose to have fewer children than those women who are paid less. Along with this hypothesis, I am assuming that a woman with a higher wage will also have more schooling than a woman with a lower wage. This assumption of course is somewhat typical for most individuals in the labor force, regardless of their sex. So with both of these hypothesis we can ultimately assume that women who have more schooling choose to have fewer children. When you think about that conclusion you might ask yourself,why “better of” women might make such a decision.

With my hypothesis, the assumption is being made that women value their schooling and income, more so then having children. This may seem somewhat rash, but there is a case to be made nonetheless. The opportunity cost it takes to raise a child instead of working can be substantial. Especially for women who have taken the time to go to school, earn their degree, and spend their own money on education already. A woman who has a degree, and is making more money than a woman with just a high school education, might have to think twice about taking the time away from work to raise a child. After all, that could add up to a lot of income she will no longer be reeling in! On the other hand, a woman who has not earned a degree, and is not making a lot of money, could see the opportunity of having a child, and think that she is not losing as much. With that said, most people see the opportunity of having a child as a great blessing, regardless of the opportunity cost involved.That fact alone will be a crucial unmeasurable variable in our error term.As an economist you have to think at the margin though, so with this hypothesis we are doing just that; while trying to account for as many variables as possible when discovering why women might make the choices they decide to make at the end of the day.

With the data we have in our research we have a total of 753 observations, and 22 different variables.Of those twenty two different variables I would like to focus on some of the specific variables that will affect our hypothesis such as,in the labor force, wage, experience squared, age, education, city, mother’s education, father’s education, kids less than six years old, kids greater than six years old,husband’s wage, and non-wife income. These variables will make up the multiple regression models we will want to look at to test whether or not our hypothesis is in fact true, or false.Using these variables in a regression function on wage will be our first step. Some obvious variables that will affect a women’s wage would be age, experience squared, education, and city. It is safe to say that these variables affect almost anyonesindividual’s income, but this regression will give us a good idea of what a women’s true wage function should look like based on our sample. So the true regression function would be “Wage = Age + Experience Squared + Education + City”. Referring back to all of the crucial variables mentioned before, I believe those as well will have an effect on a married women’s wage, so we can add them to the regression, and test whether or not they are significant. By doing so, our new regression model turns out to be “Wage = Age + Experience Squared + Education + City + Mother Education + Father Education + Kids + Non-Wife Income”. This regression will set us up for developing models to solve our hypothesis.

Some important figures to keep in mind during the analysis are the means of certain variables. Another important figure to note is that of the 753 women observed only 428 of those women are in the labor force. That is about 57% of the total women observed, which is important because we can keep in mind that there are more women working then not working. This also tells us that a large portion of women simply are not working based on our survey.When we look at the results from our regression model “Wage = Age + Experience Squared + Education + City + Mother Education + Father Education + Kids + Non-Wife Income”, we find that many of these variable come up as insignificant. We can also run this same regression for Log Wage which is the“lwage” variable instead of “wage”. When looking at both sets of results you find that the only true significant variables are “education” and “experienced squared”. You know that these variables are significant by looking at their P-Value which is shown in the “Pr>t” column of the SAS output chart on the last page (see charts 1&2). This seems unlikely that only these two variables are significant when determining a married woman’s wage, for example in theory our “Non-Wife Income” variable could prove to be somewhat significant. After all if a women has a substantial amount of income from anything other than her working herself, it would make sense that she does not have to work as much, there for she would have a lower wage. For instance her husband makes a lot of money, or possibly she has some other type of funds given to her without having to work like a large sum of inheritance.The results from our regression models influence a different test to perform, and that would be a probability test.

A probability test in SAS is called “procprobit”, and with this probability test we can decipher whether, or not, all of variables are probable, or improbable. In this test we will want to look at whether or not the specific variables used before are significant in regards to woman being in the labor force. As noted earlier only 428 women surveyed said they were in the labor force, so to run this test you want to make sure that you specify the results to those examples. You do this by making sure SAS reads the dummy variable from the survey correctly, and in the survey when answering “yes” to in the labor force, the women responded with a “1”. In ProcProbit, SAS is basing its results off the number of zeroes in the data, so you simply set the answers of “1’s” to zeroes to make sure the data is read correctly. Since the results from the regression analysis tell you which variables are very strongly insignificant, like “father education” for example, we can already assume that those variables might not be significant when it comes to a women being in the labor force. Whether, or not a woman is even in the labor force certainly affects our hypothesis though, so running a probability analysis on whether or not our specific variables effect a women being in the labor force is necessary.The results show that the variables “Education”, “Age”, Experience Squared”, “Kids Less Than 6”, and “Non-Wife Income” all have a significant effect on a woman being in the labor force (refer to graph 3). This is interesting in regards to the hypothesis, because the results show that having young children affect a woman being, or not being, in the labor force, when our hypothesis assumed the age was irrelevant. While this data is useful we still need to perform one final test to get a truly confident idea on whether or not we should reject the hypothesis; along with a simple check on the means of the data as well.

The last test to perform is a “white test”. This test will resolve issues of heteroskedasticity in the data. For this analysis we will use the variables “education”, “kids less than 6 years old”, and “kids greater than 6 years old”. Since these variables are specifically related to our hypothesis, this test gives great insight to whether or not we need to reject our hypothesis or not. The results from this test show us that the variable “education” is in fact significant once again, while “kids less than 6” and “kids greater than 6” both seem to be insignificant. The white test itself has a p-value of .0564, which means we are very close to heteroskedasticity at the 5% level, and most certainly do at the 10% level (see graph 4). We can also use the “acov” function in SAS to correct our standard errors slightly. As expected though, our variable education is still very significant even after this extra step in SAS.

For good measure at the end of the analysis I chose to separate the data into two sections, women in the labor force, and women who are not. After separating the women into those two groups, I ran a simple “proc means” function in SAS which simply gives the means of all the variables in each group. When comparing the two group means you can see that women in the labor force do have a higher “education” average, as well as a higher “wage” mean. Unfortunately the mean for both “kids less than 6” and “kids greater than 6” are about the same for each group. This small check can give us a little more reassurance that the test we have performed thus far are giving us the correct results.

With the data gathered from all of the tests which include the regression, probability, white spec acov, and mean analysis, we should reject the hypothesis that women with more education choose to have less children. We can say that “education” is a very significant variable on a woman’s “wage” though, and that more “education” does suggest a higher chance of being in the labor force, as well as having a higher “wage”.We can also note that women who are in the work force are affected by the variable of having younger children, which would be related to our hypothesis even though age was never specified. Some critical issues with the data set as a whole as well, is that the data is fairly old. It is possible that with more recent data our hypothesis could in fact prove true since women in the work force and women education levels has drastically changed in recent years. Another issue from the data is that it is not exactly specific to a group. In a different study I would prefer to look at a large sample of women with at least a bachelor’s degree, and compare it to a large sample of women who have less than a bachelor’s degree. I think that would give better insight into our hypothesis since this data did not show a very significant difference in the amount of schooling between the women in the labor force, and the women not in the labor force. Also if possible you could try to look at the husbands schooling as well in a different study since that might also prove to be significant when looking at a total family income. Since we know education has a positive effect on wage, it would be nice to compare married women who have a high wage, to women who do not have as high of a wage.

In conclusion, we found that having a higher education does have a positive effect on a married women’s wage. We also discovered that variables such as having young children, her age, experience, as well as non-wife income affect the probability of whether or not a married women is in the labor force or not.We can also reject the hypothesis that married women with more education have fewer children.

Graphs:

(1) Regression Model Results:
“Wage = Age + Experience Squared + Education + City + Mother Education + Father Education + Kids + Non-Wife Income”

(2) Regression Model Results:
“Log Wage = Age + Experience Squared + Education + City + Mother Education + Father Education + Kids + Non-Wife Income”

(3) ProcProbit (Probability in Labor Force):
“In Labor Force= Age + Experience Squared + Education + City + Mother Education + Father Education + Kids + Non-Wife Income”

(4)White Spec Acov
wage = Education Kids Less Than 6 Kids Greater than 6 / white spec acov

Proc Means:

Proc Means for “In Labor Force” and Not in Labor Force:

SAS Code:

libname prjct_19 'C:\Users\User\Documents\Kent_19';

filename MROZ 'C:\Users\User\Documents\Kent_19\MROZ.raw';

data temp19;

infile MROZ;

inputinlf hours kidslt6 kidsge6 age educ wage repwagehushrshusagehuseduchuswagefamincmtrmotheducfatheducunem city expernwifeinclwageexpersq;

run;

Procmeans;

Run;

procregdata=temp19;

model wage = age expersqeduc city;

run;

procregdata=temp19;

model wage = age expersqeduc city motheducfatheduc kidslt6 kidsge6 nwifeinc;

run;

data new;

set temp19;

depinlf = 1 - inlf;

procprobit;

classdepinlf;

modeldepinlf = age expersqeduc city motheducfatheduc kidslt6 kidsge6 nwifeinc;

run;

dataneww;

set temp19;

depinlf = 1 - inlf;

procmeans;

classdepinlf;

run;

ProcRegdata=temp19;

model wage = educ kidslt6 kidsge6 / whitespecacov;

run;

1