« More statutory interpretation from Donald Sterling | Main | A Flawed NRC Report: Prison Populations and Sentence Length, Part 2 »

Thursday, May 29, 2014

A Flawed NRC Report: Prison Populations and Sentence Length, Part 1

(This is Part 2 in my criticism of the recent National Research Council report on incarceration. Here is Part 1.)

Perhaps the most common explanation for prison growth is that sentences have simply gotten longer. It’s an understandable position, given the frequency with which we hear about long sentences for crimes that seem not to deserve them. And legislatures have spent years passing tough new laws, such as three-strike laws, truth-in-sentencing provisions, and just tough new punishments in general.

But for all its intuitive appeal, the connection between prison growth and sentence length is complicated and likely overstated. Unfortunately, the NRC report’s treatment of this issue is remarkably simplistic. The best that can be said of it is that it significantly and unacceptably understates the complexity of the issue; the worst is that it oversimplifies to the point of just being wrong.

The NRC report is unambiguous about the role of time served. In its conclusions, it states:

[I]ncreased time served is closely associated with incarceration for violent crimes and explains much of the growth since the 1980s. 

 The body of the report itself is somewhat more nuanced, arguing that longer sentences explain growth during the 1990s, and increases in admissions per arrest the growth in the 2000s. As I will argue soon in a future post, the report appears to be right about the causes in the 2000s, although its treatment of this issue is similarly dangerously oversimplified. But there are reasons to doubt its conclusions about the 1990s as well.

First, as I have argued repeatedly, it is hard to detect the effect of longer sentences being imposed, even in the 1990s. Consider the following arguments:

1. Using inmate-level data from eleven (disproportionately Northern and blue-leaning) states from the 1980s to the early 2000s, I couldn’t find evidence that sentence length played any role in prison growth (and my data included the carceral behemoth of California). The detail in my data allowed me to run simulations in which I assumed that the sentencing practices in place in the late 1980s remained in place over the whole sample period, and I found that there was almost no difference between the simulated prison populations and the real ones. In other words, the passage of new sentencing laws in the 1990s did not noticably shape prison populations. I also found that most prisoners served fairly short sentences.

2. There was a potential limitation with that earlier study: it could be that even though most inmates served fairly short sentences, a small cohort of very “durable” offenders could create a large bloc of unreleased inmates that would bolster prison populations for a long time. Using the same data but projecting forwards rather than backwards, I found that such a core did exist but (under some contestable assumptions) would not exert too much pressure on prison size.

3. Moving beyond the fourteen states used in the first two papers, I expanded my reach to all fifty states over the period 1977 - 2010. I made a simple assumption: what if I just assumed that every admission cohort in every state starting in 1977 faced the exact same release schedule (40% out within a year, 10% out within two years, etc., etc.). How would a fake prison population based on an invariant release schedule compare to the real one? Much to my own surprise, it tracks is almost perfectly, for every state.

Note that my studies all cover the years when the NRC report says that increased sentence length was at the heart of prison growth, and they don’t find it.

Now, you don’t have to agree with my conclusions: two very smart economists don’t, and the concerns they raise are ones that deserve attention (although I still stand behind my results). There are tricky methodological issues at play here, and even tricky semantic ones: if someone goes from serving probation to two years in prison, is that an increase driven by admissions or by longer sentences (from zero to non-zero)?

But an NRC report is supposed to engage rigorously with the entire literature, and here the report has completely ignored dissenting views from its conventional take. And not just random, obscure disagreement: one of my articles was published in the leading peer-reviewed law and economics journal. So if not wrong, the report glosses over—by which I mean completely ignores—significant disagreement about its conclusions.

If nothing else, just look at this figure (from here), which plot annual admissions to and releases from state prisons. If sentence lengths were getting tougher, the two lines should diverge, but except for a brief moment in the early 1990s—admittedly consistent with the NRC report’s claim—it doesn’t. And it isn’t clear that the gap is big enough to be the key explanation for that period either (especially given the other results I have derived). Clearly this issue is far more complex than the NRC's analysis suggests.

Screen Shot 2014-05-29 at 11.09.07 AM

I’ll end this post here. A second post, which will also go up today, will look at problems with the NRC’s analysis itself. But I don’t want to cross (even more that I already have) the tl;dr line.

Posted by John Pfaff on May 29, 2014 at 11:11 AM | Permalink


These are interesting observations. I would appreciate a response to the following two points. First, the parallel rising curves for admissions and releases in Figure three doesn’t preclude the argument that increased sentence length is significantly contributing to overall prison population growth. Longer terms for some offenders (e.g. violent offenders) can be offset in the data by a large influx in the number of offenders receiving shorter terms (i.e. drug offenders). Hence both truth in sentencing laws and increases in the incarceration rate of drug offenders contribute to overall population increases. Second, from a policy perspective, the focus on total admissions obscures an important distinction. According to BOJ new court commitments to prison increased (approximately) fourfold from 1978 to 2008 while parole revocations increased tenfold. From 2008 to 2012 new commitments decreased by 7% while parole revocations decreased by 40%. These data make clear that any discussion of prison population increases/decreases without considering the role of parole revocations--typically involving less serious offenders and more easily controlled by administrative authorities—overlooks an important distinction. (I’m an Emeritus Professor of Sociology at CSU Sacramento).

Posted by: Kingsnorth | May 31, 2014 10:19:14 AM

Both of these are important questions--thanks! Concerning the first, your concern is the very one that Neal and Rick (the "two very smart economists" above) raise with my paper. And it is a criticism *I* raise with my paper as well, one I try to address both in that paper and in a follow-up piece (ironically published earlier) in the U Chi Legal Forum. I can't say I've disproved the everyone-is-serving-longer-sentences effect, but I think I've at least shown that it isn't the only thing happening, and I haven't been convinced yet that it *is* the main story. The primary dataset I use, the National Corrections Reporting Program, was recently revamped for the years 2000-2012 in ways that make this issue easier to examine, and I plan to look at it more closely soon. (That said, the range of years for the improved data unfortunately does not cover the 1990s.)

As for parole revocations, first, national-level stats have to be treated carefully. California is a huge outlier--where most states' admissions cohorts are about 30% to 40% violators, California's have be about 70%, at least pre-Realignment--and has a prison population so large it can skew national statistics; BJS reports now report California separately from the rest of the country. But when I looked at parole violations in the NCRP, what I found (and this is in my ALER paper) is that (1) over the course of the 1990s and early 2000s, violations rose from about 30% of admission cohorts to about 40%, so the *change* in violations wasn't that big, and (2) parole violations and parole releases track each other closely, with releases always more than violations.

The latter point led me to think of parole violations the following way: you're in a boat filling up with water, and you have a bucket to bail out water, but the bucket has a hole in it. So as the boat fills with water (admissions), and you bail out the water (parole releases), some of the water falls back into the boat (violations). The more water, the more you bail, and the more comes out of the hole. But you would never say that the hole in the bucket is *causing* the boat to fill with water. Causation runs the other way.

Obviously that's a bit of an oversimplifcation. But it strikes me that that metaphor is more relevant than it gets credit for. This, too, is a result I can reexamine using the revised NCRP, and which I plan to do in the future. Can you give me a link to the BOJ report, though? And is that nationwide, state-specific, or federal? Thanks!

Posted by: John Pfaff | Jun 2, 2014 11:36:52 AM

Post a comment