UNIT 10: CAUSAL ANALYSIS:

EFFECTS OF EDUCATION AND OCCUPATION ON EARNINGS

In this unit we are going to ask about the effect of more education on earnings, and the effect of occupation type on earnings, and then figure the effect of education and earnings, taken together, on earning. And we will be taking up some new procedures.

Open datafile WORK9X.DAT. This file was made from US Census data by Bill Frey's group at the Population Study center at the University of Michigan. In the Command menu, do Info, and you will see the names of the variables, and how many categories each has. Then do All Marginals, to see the categories and what percent each makes up of the 58 million full-time workers aged 25-54 in the 1990 Census. While the Marginals are on the screen, notice the following:

RACELAT. NHwhite means non-Hispanic white. AllOther includes Latino and Asian Americans, and a tiny segment of Other.

OCCupations. In our previous work Profl included higher professions (doctors, lawyers) and lower ones (nurses, social workers), and Mgr included highly placed and lowly placed managers. The present classification divides jobs into Top WC (white collar) and Otr-WC (other white collar). Service is by itself, as before. And the new Lab+Farm, a large category, is rather mixed, including farm, with around 3 percent of all workers, and then manual workers of all kinds, including both the loftier Crafts, and the lowlier operators & laborers of our previous OCC8 variable.

Now run Crosstabs EDUC / EARNING pct Across. Take a look at the table on the screen. Reading up the <15K (less than $15,000 a year) column, we see a strong effect of education level. The higher the education level, the lower the percentage with this small earning level. And read up the 50K+ column: the higher the education level, the greater the percent making that much money -- and the same tendency can be seen in the 35-50 column. But what can we do with the 15-25 and 25-35 columns?

Important point: When the dependent variable (the effect), in this case EARNING, has more than two categories, it is a puzzle to read the table, and a puzzle to graph. Ideally, we nearly always want to graph the percents for just ONE category of the dependent variable. We can do that here by combining some of the higher earning categories into one new category, and some of the lower ones into the other new category. But how do we know what groupings to make?

Still in the EDUC / EARNING crosstab, do Pct diff in the Options menu. For positive categories, pick LtHS, the lowest education level, then No more. Now pick for the negative category Collgrad, and then No more.

Now you will see the same EDUC / LEARNING table as before, but underneath there is a % Diff row. Under <15K you see 32.30 -- which is the LtHS percent 38.6 minus the Collgrad percent 6.3 -- and so on across. Notice that at 25-35 the signs in the % Diff row change to negative -- meaning it is here the LtHS percent (+) starts to be less than the Collgrad (-) percent. So we can tell where to divide EARNINGS into two categories.

Now let's do it. In Options Exit out of the crosstab. Open the Modify menu. Pick Combine. You will see a list of the variables. Pick EARNING. Now, carefully, pick 25-35K, then 35-50K, then 50K+, then No more. A box invites you to name your new combined category. Type 25Kup, and Return (or OK, or in DOS, Enter).

Now open the Modify menu again, (you may have to wait while CHIP does the combining), pick EARNING, then pick <15K, then 15-25K, then No more, and name this Lt25k. If at any time in the Modify-ing you mess up, hit Cancel and start over.

Let's look at our result. In Command menu, do Marginals, and pick EARNING. You will see how our two combined groups divide the total of the workers, quite evenly.

Now Crosstab EDUC / EARNING pct across. We can use either column, so let's use 25K up. MAKE A GRAPH, (line graph!) Fig. 1, "Percent earning $25K or more, by Education. Full time workers, aged 25-54, Source 1990 Census, <WORK9X.DAT." Across the bottom, use categories LtHS through Collgrad, left to right. Round off the percents to the nearest whole percent. Next to this table, write the difference "D Collgrad - LtHS = 75-27 = 48.

Now, what effect does the job type have on earnings? Crosstab OCC / EARNING, and take a look at the result on the screen. The 25Kup percents don't go in order from Top-WC to Lab+Farm. We ran into result like that before -- some "manual" jobs earning more than some White Collars jobs. For convenience, let's reorder the Categories.

So in Option, Exit, and open the Modify menu again. Pick Reorder cats. Then pick the variable OCC, and pick, in order, Service, Otr-Wc, Lab+Farm, Top-WC. Now again, Crosstabs OCC / EARNING. GRAPH Fig. 2, "Percent earning $25K or more, by Occupation. Full-time workers, aged 25-54. Source 1990 Census. <WORK9X.DAT." Horizontal scale is Top-WC to Service. Next to or in the Table, write D Top-Wc - Service = .....round off the percents first, then subtract, and write the D for this table.

We will be using this datafile again, as we have Modified it. In the File menu, pick Save As and call the new version of the file WORK9Y.DAT. Make sure it ends up in the AI1 folder or subdirectory, along with WORK9X.DAT.

Now let's see the effect of both Education and Earnings taken at once. Crosstab EDUC / EARNING, and in Options, Control OCC. Percent across. You will get four Subtables, namely EDUC / EARNING when OCC is Top-WC, then the same when OCC is Lab+Farm, then the Otr-Wc subtable, and finally one for Service.

GRAPH Fig. 3, "Percent earning $25K or more, by education, by occupation. Full-time workers, aged 25-54. Source: Census 1990. <WORK9X.DAT." Make a Slope graph from each of the four subtables, thus a line for the 25Kup percents in each subtable. Across the bottom, the categories of EDUC. Draw lines to connect vertically the points over LtHS, then another line connecting the points of HSgrad, etc. for the other two EDUC categories. This is called a FISHNET graph. Here you can see the effect of EDUC and OCC at the same time on EARNING. Find the highest percentage in the whole graph, the point for the people with highest job, and the highest education level. Subtract from that number the lowest percentage, the one for the lowest job and the lowest education. Write on the table: Combined D [highest] minus [lowest] = >