Florida 2000 and Washington 2004

"Perhaps" they were? “....as far as we can tell?" The claim that Lichtman absolutely did <\i>not control for variables other than race is a recurring and strident theme throughout Thernstrom and Redenbaugh's dissent. Yet here we're told that it is "impossible to tell" whether he did or not. On page 19 they state that,

"The obvious explanation for a high number of spoiled ballots among black voters is their lower literacy rate. Dr. Lichtman offers only a perfunctory and superficial discussion of the question, and fails to provide the regression results that allegedly demonstrate that literacy results were irrelevant.....

Moreover, the data upon which he relies are too crude to allow meaningful conclusions. They are not broken down by race, for one thing."

(Thernstrom & Redenbaugh, 2001)

Here, and throughout their dissent Thernstrom and Redenbaugh claim that Lichtman's data and results were never provided to them (a point to which we will return shortly). Yet they are presenting us with tangible and specific criticisms of his literacy data even though they have repeatedly claimed never to have received it. How can this be? Whether they examined it or not, Lichtman's data is easily accessible from his cited sources. These datasets are among the most thorough and specific available, and all are available online. A check of the 1990 Census data he cites for Florida education level demographics reveals that it is in fact broken out by race. They insist that it wasn't, but never once are we referred to this data as Lichtman cited it and shown where it is lacking.

Having read through Thernstrom and Redenbaugh's dissent dozens of times I am unable to find any specific criticism of Lichtman's analysis beyond mere assertions of inadequacy like these. Dozens of pages are devoted to discussing the importance of controlling for variables other than race. Multiple examples are provided of studies that failed due to a failure to do this. None bore any resemblance to Lichtman's analysis or the subject race and ballot spoilage. In the end, the entire argument boils down to little more than a gut feeling that his data and methods were flawed defended with the claim that specifics of each were never made available to them.

Which brings us to the next subject.

Lichtman refused to provide the dissenters with his data and results.

We saw earlier that in his report and testimony before the Commission, Lichtman gave detailed descriptions of his data, regression methods, and his results. Thernstrom and Redenbaugh bluntly challenge this claiming that no such information was ever provided to them. They even claim that the Commission told them it was "literally unavailable" and that every word of Lichtman’s report was hearsay. On page 2 we’re told that,

"[A] request for basic data to which we--and indeed, any member of the public--were entitled was denied to us. The Commission hired Professor Allan Lichtman, an historian at American University, to examine the relationship between spoiled ballots and the race of voters. We asked for a copy of the machine-readable data that Professor Lichtman used to run his correlations and regressions. That is, we wanted his computer runs, the data that went into them, and the regression output that was produced. The Commission told us that it did not exist—that the data as he organized it for purposes of analysis was literally unavailable. Professor Lichtman, who knows that as a matter of scholarly convention such data should be shared, also declined to provide it.

Even now, five weeks after our first request, we still have not received the multiple regressions and the machine-readable data that were used in them. They are the foundation upon which the Commission's report largely rests."

(Thernstrom & Redenbaugh, 2001)

These are serious charges. Lichtman and the Commission majority are being accused of deliberately hiding their data and results--perhaps even fraudulently misrepresenting them. If true, this would undermine all confidence in the Commission's report. It's noteworthy that they restrict their accusations to machine-readable data. Lichtman's data sources were all clearly cited in his report. Thernstrom and Redenbaugh made no reference to any of these in their dissent, but it appears from this that they were aware of them. What they are asserting is that the Commission refused to supply this data and the actual regression runs in a form that would allow them to independently verified. Was this the case?

Whether the printouts they were given during Lichtman's testimony included his actual model runs is not clear. But this is irrelevant. Multiple regression methods are straightforward as are the specifics of the ecological and extreme analysis methods Lichtman used in his runs are presented in detail in his cited sources (Blaloch, 1978; Lichtman & Langbein, 1978; Lichtman, 1991; Grofman, et. al., 1994). Numerous industry-standard programs for running models like these are readily available (e.g. SHAZAM). The critical component is in the choice of variables used and the type and quality of data input to them. If this information is available, it would be straightforward to generate computer runs that would verify or refute his results and actual printouts would not be necessary.

At this point, any serious investigator would have checked Lichtman's sources to see if the data reported there wasn’t available in machine readable form. Incredibly, Thernstrom and Redenbaugh appear not to have done this.

So I did. Guess what I found.

On page 24 Thernstrom and Redenbaugh cite Lichtman's precinct level ballot spoilage data for Miami-Dade, Duval, and Palm Beach Counties as among that which was never made available to them in machine readable form.

"Dr. Lichtman devotes considerable space to a discussion of precinct-level variations of in rates of ballot spoilage for three of the Florida's largest counties. His machine-readable data was not made available to us, regrettably, despite our repeated requests for it, and neither were we provided the details of his regression analysis. We suspect that if we had been able to reanalyze Dr. Lichtman's treatment of precinct-level data, we would have found it just as problematic as his work at the county level."

(Thernstrom & Redenbaugh, 2001)

In footnote 3 Lichtman cites this data as having been obtained from the web site of Bruce Hansen of the University of Wisconsin-Madison at www.ssc.wisc.edu/~bhansen/vote/data.html. A check of that web page returns precinct level data files for unrecorded votes, undervotes, overvotes, and voter registration by race for all Florida counties. All files are available in text and MS Excel spreadsheet (.xls) or Comma Separated Value (.csv) formats across the board. From here it was straightforward to download the data files and save them to their native Excel or csv formats, plain ASCII text, unicode text, .dat, or xml formats in seconds. In these formats the data is compatible with virtually all modern operating system environments including Unix (BSD, AIX, Sun OS, HP-UX), Linux, Mainframe (OS/390), Windows, and Macintosh and can be directly read by any multiple regression or numerical analysis modeling program in existence. The xml versions can be read and analyzed by a number of web-based programming environments including Common Gateway Interface (CGI), Active Server Page (ASP), ASP.NET, and Java based web applications. It can even be analyzed in distributed program environments utilizing multiple disparate computing architectures and automated data transfer via "web services" integration environments (e.g. Vitria BusinessWare).

Total time to locate, download, and convert all of this data into each of these formats using nothing more than Lichtman's footnotes:   Under 30 seconds.

On pages 18 through 21 Thernstrom and Redenbaugh state that,

"The supposed refinements in Dr. Lichtman's regression analysis did not include using poverty rates as a variable, as far we can tell. Nor did they include measures of median family income, population density, proportions of first-time voters, or age structure, to name a few about which census data is readily available. So when the report declares that the answer to the question of whether other factors could have produced the ballot is "no," it is deceptive. In fact, Dr. Lichtman has no idea what role "other factors" like poverty may have played, because he did not take them into account in his analysis.

....the commission refused--and still refuses--to provide us the machine readable data Dr. Lichtman used in his analysis....

We have specifically and repeatedly asked the commission to provide us with the details of this regression analysis performed by Dr. Lichtman and the data on which it was based. But our requests have been denied."

(Thernstrom & Redenbaugh, 2001)

Lichtman cites the 1990 U.S. Census for his socioeconomic data (USCB, 1990). A brief search of the American Factfinder section of the U.S. Census Bureau web site reveals two Summary Tape Files (STF 1 and STF 3) containing detailed and summary data for the 1990 U.S. Census (full URL's are included in the citation). Both are presented as searchable databases where family structure, median household income, education level, education level by race, and dozens of other socioeconomic datasets are provided for sample sets as small as Census Tracts. All requested data was returned in HTML table format from which it copied and pasted directly to MS Excel spreadsheet and/or standard text editors from which it could be saved to any of the machine readable formats listed above. All search results were accompanied by print quality PDF reports of standard errors and variances in the data returned.

A few paragraphs back we saw where Thernstrom and Redenbaugh had accused Lichtman of using literacy data, including education level attained, that was "too crude to be useful" and "not broken out by race". A check of this database for Palm Beach County Census Tract 1.01 returned education level attained, school enrollment, school enrollment vs. employment status, and even school enrollment vs. armed services status--all broken out by race--for a single neighborhood 2 miles by 1.5 miles in size just north of Jupiter, FL. If this is their idea of "too crude to be useful," we have to wonder what wouldn't be.

Total time to locate, download, and convert all of this data into each of these formats using nothing more than Lichtman's footnotes:   Less than 10 minutes.

Similar results were achieved for the rest of Lichtman's cited data. In every case I was able to locate, download, and convert all of it into numerous machine readable formats in minutes--in most cases I was able to do it in under 30 seconds. All of this data was available in July 2001 when the final revision of the dissenting statement was published. With the exception of the XML conversion capability and ASP.NET framework, so were all the tools mentioned. Other investigators had no more trouble accessing Lichtman’s data than I did. Klinker accessed it immediately and was able to replicate Lichtman's results at higher correlations using larger samples of the same variables--and his results were provided to Thernstrom and Redenbaugh prior to release of the final version of their dissent (Klinker, 2001).

The fact of the matter is that the Commission's datasets and methods were easily available to anyone who truly wanted them, and Thernstrom and Redenbaugh were shown where to obtain it. They even received printed handouts of it during Lichtman's testimony before the Commission (Lichtman, 2001b). They simply didn't bother to access any of it.

Incredible as it may seem, it appears that Lichtman was accused of dishonestly suppressing his data and results simply because he didn’t deliver it to them by hand personally, and instead had the nerve to expect that they would actually read his citations and download the data for themselves... the way the rest of us did.

An independent and more thorough regression analysis refuted Lichtman's models.