15.02.2014 Views

The Benefits to Taxpayers from Increases in Students - RAND ...

The Benefits to Taxpayers from Increases in Students - RAND ...

The Benefits to Taxpayers from Increases in Students - RAND ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Analytic Approach 25<br />

but a very small percentage of families pay more than $50,000 <strong>in</strong> federal <strong>in</strong>come taxes.<br />

<strong>The</strong> few observations with very large values can dom<strong>in</strong>ate the estimation, such that,<br />

although the result<strong>in</strong>g model describes the relationship between the <strong>in</strong>dependent variables<br />

and the average value of the dependent variable, it does not describe well the<br />

relationship between the <strong>in</strong>dependent variables and the value of the dependent variable<br />

<strong>in</strong> the typical case. Us<strong>in</strong>g logarithmic or square root transformations reduces the <strong>in</strong>fluence<br />

of the outliers, so that the model more accurately describes the relationships for<br />

the large majority of cases.<br />

We compared the differences between the fifth and fiftieth and the fiftieth<br />

and n<strong>in</strong>ety-fifth percentiles <strong>to</strong> decide between the logarithmic and square-root<br />

transformations.<br />

When we use the logarithmic transformations <strong>to</strong> estimate the taxes that <strong>in</strong>dividuals<br />

with a given level of education and given characteristics will pay or the social program<br />

benefits they will obta<strong>in</strong> <strong>in</strong> each year of their life, we take antilogarithms of the<br />

projected values <strong>to</strong> convert the estimates <strong>to</strong> dollars. In do<strong>in</strong>g so, we obta<strong>in</strong> projections<br />

of the median values of the taxes and benefits. When we use the square-root transformations,<br />

we square the projected values <strong>to</strong> convert the estimates <strong>to</strong> dollars. <strong>The</strong> result<strong>in</strong>g<br />

projections are close <strong>to</strong> the median. In right-skewed distributions such as those<br />

exam<strong>in</strong>ed here, the median is less than the mean. Consequently, our approach generally<br />

underestimates the effects of education level on government revenues and costs.<br />

Critical Assumptions<br />

Throughout this analysis, we assume that the relationships we observe <strong>in</strong> the SIPP<br />

data between <strong>in</strong>dividuals’ education levels and government revenues and costs reflect<br />

the effects of educational atta<strong>in</strong>ment on earn<strong>in</strong>gs and, consequently, on tax payments,<br />

participation <strong>in</strong> and benefits <strong>from</strong> social support programs, and <strong>in</strong>carceration. <strong>The</strong>re is<br />

powerful evidence that <strong>in</strong>creases <strong>in</strong> educational atta<strong>in</strong>ment result <strong>in</strong> higher earn<strong>in</strong>gs,<br />

on average. Higher earn<strong>in</strong>gs translate directly <strong>in</strong><strong>to</strong> higher tax payments. As noted previously,<br />

higher earn<strong>in</strong>gs both limit eligibility for participation <strong>in</strong> most social support<br />

programs and are <strong>in</strong>versely related <strong>to</strong> the benefits <strong>to</strong> participants <strong>in</strong> most social support<br />

programs. Hence, it is reasonable <strong>to</strong> assume that educational atta<strong>in</strong>ment is related <strong>to</strong><br />

social program costs. F<strong>in</strong>ally, there is substantial evidence that higher earn<strong>in</strong>gs reduce<br />

the likelihood of participation <strong>in</strong> crim<strong>in</strong>al activities. Nonetheless, we cannot exclude<br />

the possibility that the observed relationships between education level and social program<br />

costs and <strong>in</strong>carceration may also reflect the <strong>in</strong>fluence of other fac<strong>to</strong>rs that affect<br />

both educational atta<strong>in</strong>ment and participation <strong>in</strong> social programs or <strong>in</strong> crime.<br />

To explore the extent <strong>to</strong> which our results are sensitive <strong>to</strong> the assumption that the<br />

observed relationships reflect the effects of education on public revenues and costs, we

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!