Measuring income of accountants relative to other occupations

[Guest]

Hello everyone,

i have two questions regarding the attached regression model. The regression want to test whether the quality of accounting education has declined. The quality of an education in terms of its observable relative economic value in U.S. labor markets.

As you can see, the dependent variable (Income) is turned into percentile ranks. The author stated:

Because many of my measures are not stable over time (income for example), I cannot attribute all variation in their raw values to changes in the quality of accounting education or the quality of accounting students. Rather than rely on raw measurements, my tests all rely on measures comparing values for accountants at a given time to concurrent values for other comparable occupations or degrees. Specifically, I transform raw values, which often cannot be compared across years, into percentile ranks, which are comparable across years, and model them as a function of a number of explanatory variables

There are two things that bother me here:
1. Where would you see potential problems with this model? I was thinking of possible onmitted variable bias because of possible important variables that may not be included in the model. Any other suggestions?

2. The paper measures the change in income of accounting employees, relative to all other types of education. Now imagine the following scenario
There are two occupations. 1: Accountants and 2. All other occupations
in t0, Accountants earn 100€ and all other occupations earn 50€ So the Income from. Accountants relative to all other occupations would be 100/50 = 2
in t0, Accountants earn 90€ and all other occupations earn 25€. The income from both occupations declined but the relative income from Accountants to other occupation is now 90/25 = 3,6. So this result would suggest that the earnings of accountants have increased, when it actually decreased, right?
I was wondering if this model is taking this scenario into account? Maybe you could help me out here. Please let me know if you need any additional informations!

Thanks in regard, Guest

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex. Also, don't post pictures. And do simplify your post the the minimum needed to demonstrate the problem. We don't need to waste our time with your long, copied, equate and list of variable definitions. If it is worth our time to help you, it is certainly worth your time to write things out.

It is "change in income" or level of income? These are very different things. It would be surprising if many of these dv's explained changes in income but not surprising if they explained levels of income.

1. Anyone can always claim there are omitted variables. There are always omitted variables (a bit of an extreme statement but probably right.) For example, sometime folks include year dummies. But the percentage formulation may somewhat alleviate the omitted year issue.

2. The model does not take that into account. Many dv's have problems as do many models. You've id'ed one here. Some would run actual income but include the average income of non-accountants or some such as a control. One approach would be to run different formulations of the model and dv and see if they give different results.

Perhaps the biggest question is whether the individuals who become accountants differ in some unobserved way from other individuals.

[Guest]

Hello Phil,
first I would like to apologize for the long post and I really appreciate it that you took your time to read it anyways and give me an answer. I will keep your advice in mind.

The author tries to figure out whether the income of accountants changed relatively to other occupations over a period of time. If my understanding of this is right he measures the level of income of accountants, relative to other occupations (in percentile ranks) and and compares the change of these relative levels over time.
The results are estimations of ß1 for three different periods (1950,1970 and 2010) and 5 different subsamples. The main results include the Change of beta 1 for each of these periods.
For example:
beta 1 estimations in the sample of all workers controlling for education:
1950 -> 0,05***
1970 -> 0,06***
2010-> 0,10***
Change 1950-2010 ->0,05***

As you probably know, I am a newbie in STATA and regression analysis but what exactly is the advantage of measuring income in percentile ranks and not with the Mincer equation for example?

Thanks again for you help, I really appreciate it!