As I've repeatedly argued, Gallup's likely voter (LV) data would be inappropriate to use until right before the election even if they worked perfectly at that particular time. But it's interesting to note that their LV data lately have not been working terribly well, compared to RVs, even on election eve, which is supposed to be when theses data really shine. That kind of undercuts their pseudo-rationale for their cavalier use of these data.

After all, David Moore of Gallup has admitted: “We simply do not know, nor can we know, which model is better during the campaign itself.” But he defends the use of LV data months and months before the election because “if it is the most accurate model just before the election, it is probably the most accurate during the campaign as well.”

Oh really. Well, let's see just how accurate these data have been recently. Here are the last 4 presidential elections and the track records of the Gallup LVs and RVs in terms of predicting the final margin of the election:

2000: RVs, +1D; LVs, +2R; actual result, +.5D
1996: RVs, +16D; LVs, +11D; actual result, +8.3D
1992: RVs, +8D; LVs, +12D; actual result, +5.5D
1988: RVs, +8R; LVs, +12R; actual result, +7.7R

So:

1. In 2000, the RVs pretty much get it on the nose and correctly call the popular vote winner, which the LV data miss.

2. In 1996, the LVs are indeed substantially better--but in 1992, the RVs are substantially closer to the final outcome. And in 1988 the RVs are pretty much are dead-on while the LVs are more than 4 points off.

That's 3 out of the last 4 elections for the don't-get-no-respect RVs.

Not too impressive. And this track record justifies subjecting us to LV data from the very beginning of the campaign and acting like the RV data are somehow inferior? Puh-leeze.

As a public service, I reproduce here the fabulous questions with which the Gallup folks produce their fabulous LV samples:

1. SALIENCE: How much thought have you given to the upcoming election for president?— quite a lot, or only a little? (“Quite a lot” or “Some” as a volunteered response score one point)

2. KNOWLEDGE: Do you happen to know where people who live in your neighborhood go to vote? (“Yes” scores one point)

3. BEHAVIOR: Have you ever voted in your precinct or election district? (“Yes” scores one point)

4. BEHAVIOR: How often would you say you vote—always, nearly always, part of the time, or seldom? (“Always” or “Nearly always” scores one point}

5. INTENTION: Do you, yourself, plan to vote in the presidential election on November (*), or not? (“Yes” scores one point)

6. BEHAVIOR: In the [last] presidential election, did you vote for (*) or (*), or did things come up to keep you from voting?

7. INTENTION: I’d like you to rate your chances of voting in the upcoming election for president on a scale of 1 to 10. If “1” represents someone who definitely will not vote, and “10” represents someone who definitely will vote, where on this scale of 1 to 10 would you place yourself?

If a voter answers each of these questions the “right” way, they get a 7, miss one and you get a 6, and so on. In practice that typically means all of the 7s—given full weight—plus some proportion of those with lower scores (usually the 6s), who are weighted down so that the size of the likely voter sample matches the projected turnout for the year (apparently 55 percent this year). All other voters are discarded from the sample.

This seems like an awful lot of work for something that (a) doesn't work that well at election time; and (b) isn't even appropriate during most of the campaign.

But Frank Newport at Gallup insists this is a "scientific" approach to take to polling. Sounds more like dogma to me.