The fault in a moving average lies in its inability to reflect a major change in the underlying reality. Consider what happened on May 2. 2011 and what that did for approval polls of President Obama. Polls from before then suddenly became suspect.
Basically, Special Forces whacked Osama bin Laden. As I recall an approval poll for Virginia went from the high 40s to the mid 50s overnight.
Or try something apolitical. Texas is prone to a phenomenon known as a Blue Norther. Before the dry and much colder air mass moves in, temperatures could be as high as the mid 80s under a clear sky even in January. So let's look at the daytime highs in the five days before the Blue Norther comes by in Dallas.
Daytime highs are 72, 68, 77, 84, 81, and 85F. A rolling average of the daytime highs isn't that silly; that's how things are typically in Phoenix at the same latitude and often similarly low in humidity. The next day the temperatures reaches 82 at 1:30. Then, all of a sudden, the wind shifts from south to north, the sky goes overcast, and the temperature plummets. At 2PM one has overcast conditions, a strong north wind, and a temperature or 55F. By 3PM the temperature is down to 39F. On the next day people awaken in Dallas to find that if they go out to get the paper they find a sub-freezing 24F.
Weather forecasters do not use rolling averages of temperature to predict the daily weather. The temperature of even two days earlier has practically no value in predicting the daily high temperature that day.
Meh. Rolling averages are generally fine for tracking polls of this sort. This is likely the first of many that will be employing this kind of methodology over the cycle. You just have to understand the trendlines, which is what these kind of polls are mainly about anyway.
The comparison I use is of the same pollster from one week to another, with recognition that events can happen that make old polls completely irrelevant. Rasmussen to PPP is probably apples to oranges unless there is some huge gap in time.
I changed my polling thread to recognize that polls, however valid, from before the Conventions have all become suspect. Assumptions that I could make beforehand (that someone up 45-38 has a significant lead) may have made sense beforehand, but as the elections get closer the 45-38 lead is not as strong as a 49-44 lead even if the 49-44 lead is somewhat narrower. Winners in a binary race should be approaching the magic 50% barrier in a binary election. (I made little change on determining how a three-way race would go).
I do not treat this Presidential race like that of 2012. In 2012 we all found Barack Obama extremely predictable. His opponent was simply predictable, offering much the same policies as John McCain. Mitt Romney ran as reality compelled him to run, seeking to cultivate mass disdain for the incumbent President in some parts of the country to spread it elsewhere. The personality of Mitt Romney was not going to matter much. A President with a scandal- free Administration, one who had an improving economy, and one whose foreign policy was effective ordinarily wins re-election. Were he running against a catastrophically-failed President, Mitt Romney would have won.
This year things are very different. The incumbent President isn't running. Both Presidential nominees faced genuine competition in the primaries. Most of our assessments of Trump vs. Clinton are qualitative.
Do Conventions matter? They certainly do! They establish themes for the upcoming election. Donald Trump has left much material for Democrats to use against him. Hillary Clinton has adopted Reagan-like themes that worked well in 1980, 1983, and 1988, maybe giving some slightly liberal twists to them so that they seem genuine enough for people to believe them. Donald Trump has offended huge chunks of America without excuse. Even more significantly, Hillary Clinton has gotten a unified Party in contrast to Donald Trump.
My gut feeling is that I see a slightly-liberal version of a Reagan-like campaign against a right-wing version of McGovern-like campaign, but I have yet to have my intuition confirmed. Polls will confirm or deny my intuition, and they will show in the form of a map.
I can't imagine a strong Trump win. Such would require him to win states that Republican nominees have not won beginning in 1992, and often by huge margins. A bare Trump win looks something like 2000 or 2004, with Trump ahead in practically all of the legitimate swing states and making some states behind the Blue (Atlas Red) Firewall look shaky. A bare Clinton win has a map looking like 2000 or 2004 with her winning everything that Democrats have lost no more than once since 1992 while picking up Colorado and Nevada together or one or two swing states (from among Florida, Missouri, Ohio, or Virginia). That would probably be like Obama 2012 less Florida. Get three of those swing states (including Florida) and start picking off states like Arizona, Georgia, Indiana, and North Carolina and a Hillary Clinton wins an election like Obama in 2008, if not a little stronger.
It's practically certain that Hillary Clinton will not win an election in the style of her husband. Should she pick up states that her husband won twice but Obama lost twice by 10% or more, then she may be winning a 45-state landslide. I doubt that she can reconstruct the Southern electorate that Carter got. I can more easily imagine her winning Texas than winning Arkansas, Kentucky, Louisiana, Tennessee, and West Virginia. It is more likely that she wins an election with a map similar to those of Eisenhower in the 1950s.
But I am ahead of myself. Some things have not changed: the Democrats have lost, perhaps permanently, the Mountain and Deep South.
