CBS coordinated with YouGov to present three “battleground state” panel polls today. The polling methodology is rather peculiar because it’s not random sample polling, but rather targeted specifically to prior respondents of same, albeit earlier, YouGov polling. They fill in the non-repeat respondents from other prior contacted buckets.
The downside to this panel-approach is it’s essentially like taking a focus group and tracking their trend over-time, filling in those who don’t show up for the next interview by reaching out to other focus groups to fill the void. This type of on-line polling is not scientific and prone to flaws and/or manipulation by the interpreter of the data.
The pollster is reaching out to a specific data-set, and in order to achieve the needed total number they need to reach out to ‘known sub-sets‘ within their data-base. In order for this approach to be useful – the integrity of the people gathering the data has to be beyond reproach.
[…] For the October, November, and December waves, all respondents from previous waves were contactedto participate. Recontact rates ranged from 42% to 53% for each state. In addition, new respondentswere selected from the YouGov panel each wave. Approximately 60% of the October wave consists of reinterviews, with the remainder coming from new additions. Approximately 70% of the November waveconsists of reinterviews from the previous waves, and approximately 90% of the December wave consists of reinterviews. (link and complex pdf methodology below)
All of that said – the first panel is within the State of Iowa and shows Ted Cruz with a large lead amid the prior contacted Iowa audience.
The second state where the panel is reinterviewed is New Hampshire where Donald Trump retains a commanding lead. The media selling point on this assessment is a “surging Chris Christie” – previously predicted narrative.
The last state included in this presentation is South Carolina. Again in South Carolina Donald Trump holds a significant lead amid the polled, repolled, and reassessed panel.
Summary: I would caution anyone to take any positive or negative data from this type of methodology – it just misses too many people, samples too few, and is subject to significant (very wide) margins of error. You’ll note the error rates of 6 to 7% on ever-shrinking samples: 927 down to 517 (Iowa), 713 down to 447 (NH), 1,224 down to 528 (SC), is very sketchy.
So sketchy are the possible deviations the panel-polling automatically creates a standard of doubt on reliability. Example: is Jeb Bush in Iowa at 2% (equivocal to Mike Huckabee) as much as people might want to rejoice at this result, it is highly unlikely. Also see Rubio who is presented at 12% (Iowa) might also be at 5% with the margin of error.
This focus group approach might be good for a media narrative – but the entire unscientific data set is essentially useless for electoral trend analysis. My .02 (And yes, even if Trump were winning in all sets I’d still say the same).
[scribd id=293729851 key=key-94owpTAcZ8Sa1W8ZGNMR mode=scroll]
Remember this tweet from CBS in August? Do you see what’s wrong with it? Same ideological presentation at play within these polling constructs…
