The paradox of lowering standards

On July 23, the Maryland State Board of Education preliminarily approved new public school discipline regulations aimed at generally reducing suspension rates as well as racial disparities in suspension rates. Public comment will be solicited on the proposed regulations, and the board will vote in early December.

Concerns that minority students are suspended several times as often as whites have lately prompted a number of jurisdictions to consider relaxing discipline standards. The approach is consistent with a near-universal perception, promoted by the Departments of Education and Justice, that stringent discipline policies lead to large racial differences in discipline rates. That perception, however, is the exact opposite of reality.

Addressing racial disparities by relaxing some standards does have a long history in civil rights enforcement, especially with respect to the racial impact of employment and other tests on which minorities generally do not perform as well as whites. That works like this.

Suppose that at a particular cutoff, pass rates are 80 percent for whites and 63 percent for minorities. At that cutoff, the minority pass rate is about 21 percent lower than the white pass rate. If the cutoff is lowered to the point where 95 percent of whites pass, assuming normal test score distributions, the minority pass rate would be about 87 percent. Thus, with the lower cutoff, the minority pass rate would be only 8.4 percent lower than the white rate. Because lowering a test's cutoff tends to reduce relative (percentage) differences in pass rates, lowering a cutoff has been universally regarded as reducing a test's racial impact.

But whereas lowering cutoffs tends to reduce relative differences in pass rates, it tends to increase relative differences in failure rates. In the situation above, the minority failure rate was initially 1.85 times the white rate (37 percent versus 20 percent). With the lower cutoff, the minority failure rate would be 2.6 times the white rate (13 percent versus 5 percent).

This pattern is inherent in the shapes of risk distributions for virtually every outcome in which advantaged and disadvantaged groups differ in average susceptibility, and it is observable in a wide range of data. Income data, for example, show that reducing poverty, while reducing relative differences in rates of avoiding poverty, will tend to increase relative differences in poverty rates.

School discipline standards operate just like test cutoffs. Relaxing standards, by increasing overall rates at which students avoid discipline (the equivalent of passing a test), will tend to reduce relative differences between the rates of avoiding discipline but increase relative differences in discipline rates (the equivalent of failing the test).

Unfortunately, almost no one studying racial differences in education understands these patterns or their implications. Indeed, to my knowledge no study of racial differences in school outcomes has understood the patterns even with respect to racial differences in proficiency and other test outcomes, much less with respect to the varied other educational outcomes.

Some researchers measure demographic differences in terms of absolute difference between rates — in the hypothetical above, 17 percentage points with the higher cutoff and 8 percentage points with the lower cutoff. The absolute difference is the same whether one examines the favorable or the adverse outcome. But absolute differences also tend to change simply because there occurs an overall change in the frequency of an outcome (though in a more complicated way than the two relative differences). Like those studying racial disparities in terms of relative differences in favorable or adverse outcomes, however, researchers who rely on absolute differences do so without the least consideration of whether the patterns they observe are anything other than the typical consequences of the general changes in the frequency of an outcome.

Actual patterns of changes in measures of differences between rates do not always conform to the frequency-related patterns. Along with changes that are solely functions of changes in the frequency of an outcome, there often occur genuine changes in the comparative circumstances of demographic groups. Society's interest is solely in identifying these genuine changes and understanding why they occurred. But one cannot do that without recognizing the ways that measures tend to be systematically affected by the prevalence of an outcome.

In any case, in Maryland, as in other jurisdictions, those expecting to see reductions in racial differences in suspension rates following the relaxing of discipline standards may be in for a surprise.

James P. Scanlan is a lawyer in Washington, D.C., who specializes in the use of statistics in litigation. His email is

Copyright © 2021, The Baltimore Sun, a Baltimore Sun Media Group publication | Place an Ad