in the last election in canada, we had consistent polling that undecideds were substantial and they would not pick stephen harper. this created polls that exaggerated his position through most of the race, until the undecideds came down. i don't see any polling like that right now - both candidates are very unpopular.
you can read that two ways.
1) all of the undecideds will vote for johnson. it's a lot harder to assign all of the undecideds to a third party candidate in a two-party system than it is to assign them to the natural governing party (the liberals) in a multi-party system. it could happen. bad assumption.
2) the undecideds might distribute.
now, i hate this. you do this when you don't have any other data. it's a vanilla assumption - usually wrong. i'm going to do it anyways...
the rcp average is pinning 9% undecided. distributing...
clinton: 42.2/91 = 46.3.
trump: 37.8/91 = 41.5
johnson: 7.9/91 = 8.6
stein: 3.1/91 = 3.4
this most closely matches the 1996 results for johnson/perot...
clinton - 49.2
dole - 40.7
perot - 8.4
so, if you assume that the undecideds will distribute then the wild card in using 1996 to predict 2016 is stein. if you deliver stein to clinton, it's very close to 1992. if you don't, it could hinder her in close races.
but, what if they all go third party? then the adjusted rcp average is:
clinton: 42.2
trump: 37.8
other: 20
look at 1992:
clinton: 43.0
bush: 37.4
perot + other: 18.9 + .7 = 19.6
whether you're looking at '92 or '96 as the better predictor is going to depend on how you interpolate undecideds. so, go get me more data on undecideds.
this is not yet predictive though, either. there's too much volatility, which is very different than the last election in canada, where the conservatives had flatlined for months and nothing moved them. clinton's graph could still careen out of control, producing a landslide.
but, it's what we've got right now.
as mentioned previously, you also have to make some smart adjustments. these are national polls. so you have to distribute them, too.
but it should give an idea about which states are in play and which aren't.
1996..
States where the margin of victory was under 7% :
Kentucky, 0.96%
Nevada, 1.02%
Georgia, 1.17%
Arizona, 2.22%
Tennessee, 2.41%
Montana, 2.88%
South Dakota, 3.46%
North Carolina, 4.69%
Texas, 4.93%
Mississippi, 5.13%
Indiana, 5.58%
Florida, 5.70%
South Carolina, 6.04%
Missouri, 6.30%
Ohio, 6.36%
North Dakota, 6.81%
Alabama, 6.96%
1992...
States with margin of victory less than 7%
Georgia – 0.59%
North Carolina – 0.79%
Ohio – 1.83%
Florida – 1.89%
Arizona – 1.95%
Montana – 2.51%
Nevada – 2.63%
Kentucky – 3.21%
Texas – 3.48%
South Dakota – 3.52%
Louisiana – 4.61%
Tennessee – 4.65%
Kansas – 5.14%
Iowa – 6.02%
Indiana – 6.12%
Alabama – 6.77%