Since the beginning of May a flurry of different polls have shown us different results. The results range from Romney having a 49%-44% lead in Rasmussen’s daily tracking poll to Reuters giving Obama a 49%-42% edge. I does bear mentioning here that Romney holds a 47%-44% lead in Gallup’s daily tracking survey. Over the weekend a Politico/Battleground poll survey showed Romney ahead 48%-47% while an International Businesses Daily survey found Obama ahead 46%-43%. At the beginning of the month a Democracy Corps. survey found Romney and Obama tied at 47%.
Looking that these polls three obvious things immediately show. First, in all the surveys that use Likely Voter (LV) models (RR, Gallup. Politico and Democracy Corps) Romney does the best. In the surveys that use Registered Voter (RV) models (IBD, Reuters) Obama does the best. The reason why is very simple. Across the country there are more registered Democrats than Republicans. But on election day the turnout of partisan voters is never equal to the national numbers. Whereas RV models simply survey registered voters Likely Voter models use voter screens to get the appropriate sample of what a pollster expects the election day electorate to look like. Each pollster’s likely voter screen is unique. For example in 2010 Gallup used a seven question survey to weed out who was most likely to vote and thus be included in their high and low LV turnout models. Other pollsters, such as PPP (D) use much looser screens. So since there are more registered Democrats than Republicans a RV model is likely to lean more to the left than an LV model which reflects past turnout trends, among other things.
Second, Romney has consistently done better in the Gallup and RR tracking surveys than in polls taken over a 3-5 day time period. Tracking polls can vary in size and they usually poll about 1,000-1,500 people a day. Tracking surveys because they roam different groups every day are prone to more movement than a regular survey but Romney’ support has been consistent in the tracking surveys. Romney has held a lead as large as three in the Gallup tracking survey while Obama topped out at 7% and has fallen since then. Obama has been up by as much as three points in the RR tracking survey while Romney current enjoys his biggest lead of five points.
Thirdly, the sizes of the samples vary. Romney has done best in samples that are larger (1,000 or more) while Obama has done better with samples that are smaller (1,000 or less). These sample sizes happen to coincide with the voter model used as well. Whether this really means much or not is debatable. While larger sample sizes doing better for Romney could indicate he has better support than Obama the fact they are also LV models and thus more conservative could be at play as well. By the same token Obama’s campaign could take some good news from the fact the RV models in which he leads are more realistic of the actual electorate today and who they back.
These three factors all are important but they miss the biggest differences in each survey. As discussed above LV models differ substantially from RV models. LV models are always to a degree modelled on past turnout levels and that is where the biggest difference in the samples becomes clear. Take for instance the difference between the Politico survey and Reuters survey. In the Politico Survey a whopping 58% of the sample identifies as conservative or somewhat conservative. Furthermore only 20% of the Politico LV survey is minorities compared to 78% white. That is almost the racial composition of the 2010 midterm electorate. The Reuters poll surveyed only partisan preferences and found 47% were Democrats or leaning Democrats and 38% Republican or leaning Republicans. Only15% were pure independents. Yet the Politico survey found 40% identified as Republican and 44% Democrat.
In 2008, an extremely left leaning presidential year the electorate was made up of 39% of Democrats, 32% of Republicans and 29% of independents. Yet in 2004, a better environment for the GOP the electorate was% Democratic, 37% Republican and 26% independent. In 2000, a neutral environment 39% of the electorate Democratic, 35% Republican and 26% independent. So the Reuters survey more closely resembles the partisan composition of the 2008 electorate while the Politico survey more closely resembles a neutral political year 2000 electorate. Considering the state of the nation it is more likely the electorate will look like 2000 or 2004 than 2008.
It is safe thus to deduce from these surveys that if the electorate resembles the electorate of 2000 the election will be close. If it more resembles the electorate of 2004, or even the electorate of 2010 Romney should win. And if by some chance it resembles 2008 Obama wins relection easily. It is far more likely however the electorate looks more like the LV models some pollsters are currently using and if so that is good news for Romney. Especially considering the Obama campaign was banking on increased minority turnout not from 2010 but from 2008, and that saw unprecedented minority (due to African-Americans) and youth turnout. Until election day though we must continue to try to deduce what the 2012 electorate will look like.