I recently commented on the many sources of variability in the polls. After taking a little closer look at the polls used in Texas, I find it’s hardly a surprise that they are predicting divergent outcomes.
http://www.pollster.com/blogs/why_so_much_volatility_in_texa.php
Just a quick comment — one of the things that has bedeviled pollsters is the change in communication technology. Whereas ‘land lines’ were the rule not long ago, many younger voters now rely exclusively on cell phones. Since many polling firms don’t call cell phones, the result has been a dramatic undersampling of that demographic. When younger voters turn out in greater than normal numbers, this can skew results, and particularly when they vote disproportionately for one candidate over another.
perhaps the Republican gerrymandering of congressional districts will be a significant factor in the outcome. There’ll be a surprise.
Very interesting.
On a slightly different issue — the hybrid primary/caucus system should yield some interesting results. Given this, and the close recent polls, I think the final delegate yield will displease the Clinton campaign.
The following article from Mark Blumenthal of Pollster explains why primary polling is less exact than general election polling.
http://www.pollster.com/blogs/the_problems_of_primary_pollin.php
One of ideas presented is the “party heuristic.” Voters do not have well-defined preferences (unlike you and me, perhaps) regarding intra-party candidates, which results in a fluid situation in which allegiances can frequently switch.
I see that pollsters are claiming that they’ve solved the cell phone undersampling problem, but I’m still skeptical since there are also reports that indicate just the opposite.
My last experience in polling was in 1998 — a virtual lifetime ago in terms of communications technology — and we were clearly not reaching a representative portion of the population who were cell users at that time. Increased cell usage has prompted efforts by pollsters to reach this population, but it’s doubtful that they’ve actually solved the problem. Raising the weight on cell users helps compensate for this problem, but it doesn’t address the measurement problem. A more accurate population parameter combined with poorly measured preferences can merely produce further skewing.
Moreover, since most statistical models don’t account for new voters particularly well, because of the fact that they don’t fall into the "likely" category and often aren’t included in voter lists, there is an obvious undersampling problem with this demographic. This is compounded by cell usage. And when they tilt towards one candidate, as they currently do, the problem is further compounded.
One other factor has resulted in under/overrepresentation — the relative efficiency of recruitment & GOTV efforts. Considering that the Clinton campaign only opened field offices in Texas around ten days ago, it’s not hard to see why they might have trouble identifying and turning out their supporters when the primary/caucus occurs.
The following article discusses some of the aforementioned points.
http://www.mywesttexas.com/site/news.cfm?newsid=19317226&BRD=2288&PAG=461&dept_id=475626&rfi=6
Here’s an article from Mark Blumenthal of Polling.com about telling a good poll from a bad one. It’s easier to find the issues that are disagreed upon then answer that question, however.
http://news.nationaljournal.com/articles/mysterypollster/2008/022808.htm