Monday, July 21, 2008

Small effect size- large effect

There has been a lot of talk lately about how the absence of cellphones confounds polling data as these numbers are not often included in polls (oddly enough, it is against the law to random digit dial a cellphone via computer). Thankfully, Pew has some data out which shows how great a difference in polling there is if you include cellphones.

In their survey, they found that with cellphone users included in the general sample Obama leads by 48% to 40% vs. 46% to 41% in the sample w/o cellphones. So, there is a 3% swing of support once we include voters who do not have landlines.

Yet in reading some of the commentary on this subject (e.g. here), it seems that people don't see cellphones as having a big effect. Now, I will admit that the number is not very large- but in terms of assessing its effect, I would say that it is considerable. This type of effect comes up repeatedly in studying the mass media (for example- media violence), where small effect sizes are ignored because they are numerically small but not substantively small. A shift of 3% in the general voting public (if we foolishly assume that cell phone voters are evenly distributed) would mean that Obama wins by almost a landslide- as Missouri, Nevada, Indiana, Montana and Virginia would all move into his column from the narrow McCain column. At the very least, I would venture to guess that this almost guarantees that Obama will win at least two of these states- which would mean a very comfortable win.

No comments: