Sunday, May 12, 2013

More on Conviction Rates in Tax Cases (5/12/13)

I have had several blogs on statistics, sometimes quoting the famous quote attributed to Benjamin Disreali that "There are three kinds of lies: lies, damned lies, and statistics."  Those various blogs may be viewed under the key word statistics below, here.

In some of the blogs I have questioned DOJ Tax's claims over the years of a 95% conviction rate in the tax prosecutions it brings.  Today, I address that issue again.

This past week, the IRS released its Annual Business Report on Fiscal Year 2012 National Operations, here.  That report has some statistics for 3 years that are consistent with the IRS data book, Table 18 which may be accessed here for the years back to 1995.  More about that in a minute.

On p. 4 of the 2012 Annual Report, the following statement appears:
Conviction rate is the percentage of convictions compared to the total number of convictions, acquittals, and dismissals. The conviction rate for FY 2012 is 93.0%, 0.3% more than the FY 2011 rate (92.7%).
This claim of 92.7% for 2011 and 93.0% for 2012 is not materially different from DOJ Tax's claims of a 95% conviction rate.  So, I will refer to the claims as a 95% conviction rate claim.

My first reaction was that perhaps I have been wrong to question such a high conviction rate, since, although not backed with data, the IRS's indicated methodology for calculating the conviction rate seemed sound.  But, being the skeptic, I decided to dig deeper.  I therefore offer the following analysis from the IRS's own detailed data for a longer number of years to see if I could get to the same or roughly the same indicated conviction rate.  My analysis below summarizes the more detailed analysis in a spreadsheet I offer for download here (a zip file).

I first attempted a limited analysis, analyzing legal source tax prosecutions over the period from 2005 through 2010.  According to Table 18 for each of those years, the following is the aggregate raw numbers of prosecutions and convictions for legal source income cases:

Prosecutions (Indictments, etc.)    6,713
Convictions                                 5,649
Conviction Rate                           84.2%

The number of prosecutions generally rose over the years, so that the lag between prosecution and conviction might account for some of the difference between the 95% claimed rate of conviction and the statistical data presented in Table 18, but it surely would not account for all of it or perhaps not even most of it.  If it does not account for the difference, then there must be something I am not accounting for.  I don't know what that is.  But then, I thought, the 95% claimed rate is for all prosecutions rather than just legal source.

I then said, well, I need a broader set of numbers to test the 95% conviction rate claim.  That number could be found in the IRS data book Table 18 for each year, but a reasonable proxy is found in an IRS web site called Enforcement Statistics presenting IRS CI Investigation Data, here.  This page presents select data for years 2003 through 2013.  I chose the same years for my analysis - 2005 - 2012.  That page does not indicate the number of convictions, but it does indicate the number of sentencings which should roughly equate to convictions, with a somewhat greater time lag between prosecution and sentencing.  But, over a long time frame (2005-2012 should be sufficient) and subject to an increasing number of prosecutions that might cause the indicated conviction rate to be a little lower, here are my results:

Prosecutions (Indictments, etc.)    9,908
Sentencings                                 8,678
Conviction Rate                           87.6%

So, back to the claimed 95% conviction rate.  Several years ago when I asked DOJ Tax to explain, DOJ Tax declined to explain.  I have not asked the IRS to explain, but as noted above the IRS says that it calculated its 93% 2012 conviction rate as "the percentage of convictions compared to the total number of convictions, acquittals, and dismissals."  That sounds like the right way to do the calculation, but unfortunately I can't find reported statistics on acquittals and dismissals to verify the 93% claim.  All I could find are the statistics noted above which, when analyzed, do not seem to support the claim.  Maybe the statistics I present are consistent with the claim.  Maybe not.

I realize that there may be differences in DOJ Tax's data set and the IRS's data set.  DOJ Tax may prosecute some cases that have not been investigated by the IRS and thus  might not appear in the IRS data set.  But, even so, DOJ Tax would have to have a significantly greater conviction rate in the cases not investigated by the IRS than 95% in order to move the overall statistics upward from the IRS's statistics.

Also, the 95% claim if made in the manner the IRS describes it accounts only for convictions, acquittals and dismissals in the year involved and hence eliminates any distortion from the lag time from prosecution to the disposition event (conviction, acquittal or dismissal) that is inherent in the data I analyze.  So, it would be interesting to see the IRS data and, of course, the DOJ Tax data which might use the same methodology.  And, even if the 95% rate is correct, the analysis I have above would suggest that the conviction rate in legal source crimes is probably less than the average rate (meaning the illegal source and other financial crimes convictions are probably more than the average).

If any reader is aware of the data supporting the DOJ Tax and the IRS conviction rate claims, I would very much appreciate hearing either by comment to this post or by email to


Jack Townsend

1 comment:

  1. i think it may make the Tax Deed a bit difficult or may be easier for a while.i actually have no idea about it.thanks for sharing this post here with us.keep it up.all the best.


Please make sure that your comment is relevant to the blog entry. For those regular commenters on the blog who otherwise do not want to identify by name, readers would find it helpful if you would choose a unique anonymous indentifier other than just Anonymous. This will help readers identify other comments from a trusted source, so to speak.