Mickey Kaus comes down hard here on Henry Brady’s most recent statistics regarding the extent of unintentional undervotes caused by punch cards, but other preliminary analyses have reached the same conclusion:
(1) Michael McDonald’s updated turnout analysis, which notes, among other things: “Interestingly, even though voters who voted by touchscreen and punchcard were very similar in their preference for recalling Davis, 1.4% of touchscreen voters did not vote in the recall while 7.1% of punchcard (and 5.0% of optical scan) voters did not have a recorded vote.”
(2) Steven Hertzberg’s preliminary analysis linked here yesterday, which had a 7.7% rate for punch card undervotes compared to a 2.3% rate for non-punch card undervotes.
(3) Mark Blumenthal’s analysis (passed on to me by Mickey Kaus), which has the advantage (like McDonald) of incorporating exit poll data. Unlike McDonald, Blumenthal has exit poll data that breaks it down by voting system.
Blumenthal, responding to Kaus’s criticism of Brady, wrote to Joe Lenski to get more exit poll data. Here is Blumenthal’s analysis:

According to the totals posted on the Secretary of State website, 4.6% of the ballots cast (384,976 of 8,374,681) lacked a valid vote on the recall question. According to the exit poll, 2.6% of the those voting intended to skip the recall question. This the exit poll suggest that about 2% of the ballots cast were “lost.”
The exit poll also allows us to look at this difference by type of voting equipment. As the table shows, punch cards accounted for virtually all of those “lost” (I got the percentages of missing votes from the AP story):







Percentages	Punch card	Touch Screen	Optical scan
a) Actual missing votes	6.3%	1.5%	2.7%
b) Intended non-votes (exit poll)	2.9%	1.4%	2.5%
Estimate of “missing vote” (a minus b)	3.4%	0.1%	0.2%



So where voters used punch cards, the error rate is 3.4%, where voters used touch screen or optical scan it was less than 0.2%.
Multiply the number of missing votes cast in punch-card counties (297,775 according to AP) by 3.4%, and you get an estimate of about 160,000 lost votes. A little less than Brady’s estimate, but not much.
Some other relevant findings:
The exit poll also asked its respondents about problems experienced with voting equipment. Roughly one in ten had either “serious” (2%) or “minor” (7%) problems. As a whole, this group voted 60% to 40% against the recall. Given this finding and the size of the group, Lenski concludes, lost votes from this group “may have slightly increased the recall’s reported margin of victory but in no way affected the overall outcome.”
Finally, note that according to the exit poll, Bustamante’s support was 36% among those using punchcards, 34% among those using other methods.

I’ll have more on what these numbers teach us, and other lessons for election law from the recall, in a Findlaw column out Monday.