In October 2011, the National Education Policy Center (NEPC) of the School of Education of the University of Colorado Boulder issued a report styled “Discipline Policies, Successful Schools, and Racial Justice.” The report addressed nationwide differences in suspension rates of white and black students (as well as between disabled and nondisabled students and between male and female students), and maintains that increased suspension rates lead to increases in racial differences and that racial differences do not result from differences in the severity of conduct. Most important given the points made on the main Discipline Disparities page as to why more stringent discipline policies will tend to result in smaller relative differences in discipline rates than less stringent ones, the report shows increases in relative differences between black and white discipline rates during period of general increases in discipline rates. The report also addressed other issues relating to the utility of out-of-school suspensions, maintaining, inter alia, that more frequent suspensions did not did not improve the school environment.
Several aspects of the report may eventually receive attention here. For the present, however, I merely discuss the data on changes in racial differences in discipline rates during the periods of general increases in discipline rates.
The key data are found in Figure 1 at page 4. The figure shows that between the 1972-73 and 2006-07 school years, the black out-of-school suspension rate increased from 6% to 15% while the white rate increased from 3% to 5%.
Initially, I note that, for reasons discussed in Section A of the NEPC Colorado Study sub-page of the Discipline Disparities page, examinations of differences in discipline rates are better carried out on the basis of rates of experiencing a certain type of discipline or greater (e.g., rates of experiencing out-of-school suspension or expulsion) rather than rates of experiencing some intermediate level of discipline. But the numbers of expulsions are sufficiently low that it is very unlikely that the results would differ with respect to the key issues discussed here if one examined rates of experiencing out-of-school suspension or expulsion rather than out-of-school suspension alone.
With respect to the changes in rates in Figure 1, the report notes (at 5):
The data show substantial increases for students of all races, with a growing racial discipline gap. Specifically, K-12 suspension rates have more than doubled since the early 1970s for all non-Whites. Concurrently, the Black/White gap more than tripled, rising from a difference of three percentage points in the 1970s to over 10 percentage points in 2006, when more than one out of every seven Black students enrolled was suspended at least once.
In observing that the gap between blacks and whites more than tripled, the report makes clear that it is examining the absolute difference between rates rather than the relative difference that is the common focus. For reasons explained in the Introductory Section to the Scanlan’s Rule page, in the discipline rate ranges at issue, factors related to the shapes of normal distributions will tend to cause general increase in rates to be accompanied by increasing absolute differences between rates.
The key aspect of the pattern in Figure 1, however, is that, whereas when an outcome becomes more common the distribution-related forces tend to cause the relative difference to decrease, in fact the relative difference increased. The black-white ratio increased from 2.0 (6%/3%) to 3.0 (15%/5%). The question then arises as to how this pattern squares with the pattern described on the main Discipline Disparities page where more stringent discipline policies should be associated with smaller relative differences in discipline rates.
One possibility, though I think not the most likely one, involves the fact what I commonly discuss in terms of a pair of distributions in fact is comprised of multiple pairs of distributions as to different phenomenon. Thus, for example, suppose that the distributions of whites and blacks differed much more with respect to the types of conduct to which zero tolerance policies applied than to other types of conduct. That could lead to larger overall relative increases for blacks than whites even where, with respect to particular types of conduct, the relative differences decreasing.
As noted, however, I do not think phenomena of this nature are the most likely (or most important) explanation for larger relative increases in black rates than white rates (though such phenomena may play a role). But I think it important to keep the existence of multiple distributions in mind in consideration the arguments made about the wisdom of zero tolerance programs. For example, those arguing against zero tolerance policies cite both the large proportion suspensions for seemingly minor offenses comprise of total suspensions and the pattern whereby severe discipline makes students experiencing such discipline (especially black students experiencing such discipline) more likely to become involved in the criminal justice system. Relaxing of discipline standards may substantially reduce the suspension rates for seemingly minor offenses while leaving unchanged the rates of severe discipline for the type of conduct where such is strongly associated with the later acquiring of a criminal record. See Table 1 of the Los Angeles SWPBS sub-page of the Discipline Disparities page, which shows how the relaxing of discipline standards caused a larger increase in the relative difference between black and white discipline rates than would be expected due solely to distribution-related factors.
Whether or not the phenomenon described in the paragraph above had a large role in causing the relative difference between black and white suspension rates to increase during the period covered in Figure 1 of the NEPC national study, probably a larger role was played by the fact that, for whatever reason, racial differences in discipline, properly measured, were in fact growing during the period. The method for measuring the difference in the circumstances of two groups that is not affected by the overall prevalence of an outcome (as discussed in the Discipline Disparities page) shows that the difference between means of the underlying black and white distributions was .33 in 1972-73 and .61 standard deviations in 2006-07.
Those who might instead think that patterns like that shown in Figure 1 demonstrate that there is no basis for believing that increasing the stringency of discipline policies will tend to reduce relative differences in discipline rates (or that relaxing discipline policies will tend to increase relative differences in discipline rates) should consider the following. Any statistician who gives the matter a little thought certainly will concur that lowering test cutoffs will tend to reduce relative differences in pass rates but increase relative differences in failure rates. The conceptual basis for that belief is irrefutable. But suppose that one observes in a particular situation that lowering a cutoff does not increase the relative difference in failure rates (or does not reduce the relative difference in pass rates). It would hardly be sensible to regard such showing as refuting the soundness of the expectation that reducing cutoffs will tend to increase relative difference in failure rates. Sensible observers, rather, would conclude that there was some irregularity in the distributions or that the relationship of the distributions of those taking the test with the lower cutoff differed from that with the higher cutoff. And they would continue to believe that lowering cutoffs will tend to increase relative differences in failure rates and act according to such belief, albeit keeping in mind the instance when that did not occur.
The same holds with respect to the pattern in Figure 1 of the NEPC report. One still should expect that relaxing discipline standards will tend to increase relative differences in discipline rates.
In considering this issue, moreover, one should also be mindful that the principal thing society usually cares about is whether the circumstances of two groups differ more in one setting than another. Ignoring the way standard measures of differences tend to be affected by the overall prevalence of an outcome makes it impossible to draw sound inferences about that matter. The National Center for Health Statistics some years ago decided that it would not make an effort to determine ways in which health and health care disparities might be larger in one setting or another while taking overall prevalence into account and that it instead would simply measure all disparities in terms of relative differences in adverse outcome rates. As a result, nothing that it has said about the ways the comparative size of health and healthcare disparities has been of value.