Online Data Analysis Textbook

 Here's a link to an online political data analysis textbook I put together last spring: 

Feel free to use it if it works for you.

DeSart/Holbrook Model

In case you missed it before, I'm posting a link to the DeSart/Holbrook forecasting model . Props to Jay DeSart, who has kept this updated over the years.

Here's the bottom line, if you don't want to read the deets (but you should read the deets):

Convention Bumps Revisited

It's hard to say what we should expect in terms of polling gains from the COVID-era, remote nominating conventions over the next few weeks.  On one hand, the well-choreographed convention floor "show" will be missing, potentially robbing the both Biden and Trump of whatever benefits there are from the spectacle nature of these sorts of events.  At the same time, the candidates are still the beneficiaries of four straight days during which they probably will dominate media coverage, including two-hours of prime-time television coverage each night of the convention.  So it may well be that the impact of the conventions in 2020 are not terribly different than in other years.  We'll know more about this in a few weeks.

In the meantime,  I thought it would be useful to review the extent to which the convening party candidate receives a bump in the polls following their convention.  I've covered much of this in previous posts so I'll just briefly review a few key points here.

The figure below summarizes the size of convention bumps from 1964 to 2016 (Red=Republican, Blue=Democrat).  The convention bump is measured as the percentage point change in the convening party's share of the two-party vote, comparing polls taken between six days and two-weeks prior to the convention with polls taken during the seven days following the convention.  Note that this is a short-term measure of the convention bump and does not say anything about the rate of decay in the weeks following the convention, and different calculation methods (dates covered, polls used) could produce different results. 

 

There is a lot going on here but there are a few takeaway points.  

1.  Candidates generally get a bump of some sort.  The size of the bump is highly variable but virtually all candidates leave their convention doing better in the polls than when they went into the convention.

2.  The size of the convention bump does not predict the overall winner very well.  Just ask Presidents Goldwater,  Mondale, Dole, or Gore, all of whom had bigger bumps than their competitors.  One of the reasons for this is that candidates running way behind in the polls have an easier time gaining ground during their conventions.  For instance, in 1964 Barry Goldwater was so far behind in the pre-convention polls (averaging 21% of the vote) that it was easy for him to improve his standing by thirteen points during the convention, though he still never come close to being competitive.  At the same time, Lyndon Johnson went into his convention with 69% of the vote in pre-election polls and left the convention with no bump but still with a substantial lead.  

3.  Convention bumps aren't what they used to be (see figure below).   Prior to the 2000 election, convention bumps averaged more than six points, but that has fallen to just over two points from 2000 to 2016.  One potential explanation for this change lies in the scheduling of conventions. The 2008 and 2012 conventions were held in late August and early September, and from 2008 to 2016 the conventions were held on back-to-back weeks.  The norm in other years had been to hold the conventions in late July or early August and to separate them by two to three weeks.  What is probably most important here is holding the conventions on consecutive weeks, which means that the convention messages end up overlapping and may cancel out each other.  Another potential explanation lies in the increased polarization of the electorate.  It is possible that partisans are so much more committed to their candidates now than they were before, and there is a much smaller persuadable electorate that can be influenced by events like the nominating conventions.

So what does all of this mean for the 2020 convention bumps?  One of the the key features of the 2020 conventions  (besides the obvious, COVID-inspired changes) is that they follow the recent scheduling trend of back-to-back convention weeks, so this might limit the size of the bumps.  In addition, since both conventions are being held in late August, the pool of persuadable voter who have not already made up their minds could be fairly small.  Finally, one thing that might limit Biden's bump potential (and hint at a larger bump for President Trump), is that he already has a substantial lead in the polls and there might not be much room for his lead to grow.  If Biden does manage to get a substantial bump, it may mean that his vote potential is higher than his current polling level. 

Heightened Demographic Divides

Last week I posted about some of the findings from Alerted States, focusing on the growing educational divide among the states.  In that post, I speculated that the nature of the the 2016 presidential election was likely to contribute to even sharper demographic patterns to political support in the states.

I'll have more to say about this later, but here are a few preliminary findings that confirm that state demographic characteristics were more closely tied to state-level outcomes in 2016 that at anytime in the past several decades:


Education

Correlation=.77 (previous high (2008) =.70)

Immigrant Population

Correlation=.69 (previous high (2012)=.60)

 Religion

Correlation -.79 (previous high (2012)=-.73)

Across the board, these and other demographic and cultural indicators are more closely tied to state outcomes now than at anytime in the past several decades. More importantly, these relationships continue a pattern of increasingly strong connection between state characteristics and state outcomes since the 1992 election.  Again, I'll have more to say about this later.

The Education Gap

The education gap in candidate support has been getting a lot of attention during this election cycle, including in a Toni Monkovic had the other day at The Upshot.   I take up this issue in my new book,  Altered States, where I examine the impact of state-to-state differences in education and other demographic characteristics on presidential outcomes in the states. The educational divide in politics is not new, but it has been growing over time, in part in response to increased ideological divergence among party elites, and it currently stands as one of the most important explanations of party success in state-level presidential outcomes.

The figure below illustrates the changing nature of the relationship between education and presidential outcomes in the states.  The dependent variable (vertical axis) is the Democratic share of the statewide two-party vote and the independent variable (horizontal axis) is the percent of the citizen voting age population (CVAP) with advanced degrees.  This scatterplot matrix clearly shows the increased salience of education to presidential outcomes in the states over time.  Although there is a modest relationship in 1972 (not coincidentally, a year with ideologically distinct candidates), there is essentially no connection between  education and presidential outcomes from 1976 through 1988; then, beginning in 1992, the relationship grew stronger with each successive election until it peaked in 2008, and then receded very slightly in 2012.

The evolution of this pattern of relationships is summarized more clearly in the figure below, which plots the partial correlations¹ for the relationship between education level and Democratic support in the states over time.  Again, there is a striking pattern of increased importance over time.  

The Relationship between Level of Education and State-level 

Presidential Outcomes, 1972-2012

The evolving relationship between education and state-level outcomes in presidential elections reflects a tendency found across multiple measures of state-level demographic and political characteristics, including occupational status, religiosity, partisanship, ideology, and others.  

 

Finally, changes in education levels are connected to changes in support for Democratic presidential candidates.  The figure below examines changes in the average levels of both variables from 1972-1980 to 2004-2012, showing a clear positive relationship between increases in education levels and increases in Democratic support.  Democrats made their greatest gains among those state with the greatest educational grains and suffered their greatest losses among those states with the smallest gains in level of education:

Change in Centered Democratic Vote Share and Change 

in % of CVAP with Advanced Degrees

Note: Changes in both variables are measured from 1972-1980 (averaged) to 2004-2012 (averaged). 

Adapted from Altered States.

¹The partial correlations control for southern regional effects during the Carter (1976, 1980) and Clinton (1992, 1996) candidacies, as well as presidential and vic-presidential home state advantages.

State Polls Then and Now

There is some talk these days about the shifting Electoral College map in 2016, based mostly on Clinton's poll numbers in states such as Arizona and Georgia.  One way to get a sense of how things are changing is to compare the most recent polling averages to the polling averages for the same states at this point in the campaign four years ago. The figure below compares the current (August 9, 2016) Pollster estimates of  Clinton's share of the two-party polling preference to Pollster's estimates of Obama's share of the two-party polling preference for the same states at this point in the campaign four years ago. This figure shows important signs of both continuity and change.

Continuity.  First, in terms of relative support, the picture is one of continuity: generally, the same states that gave Obama the greatest support in 2012 are still Clinton's biggest supporters, and those states that gave Obama the least support are also giving Clinton her lowest levels of support (the correlation between the two years is .87).  The bluest states from 2012 will still be really blue in 2016 and the reddest states will remain pretty red, with the exception, perhaps, of Utah, which may take on a pinkish hue.

Note: this figure only includes states for which there were Pollster estimates estimate for this point in time in both 2012 and 2016

Change. While the relative positioning of the states vis-à-vis each other has not changed much, there has been a fairly uniform shift among the states in Clinton's favor, reflecting her standing in national polls: across the nineteen states in the figure above the average level of support for Clinton in 2016 is roughly 2.4 points higher than the level of support for Obama at this point in the campaign, and  Clinton is doing better than Obama did in almost all states (states above the line of equality are states where Clinton is outperforming Obama at this point in the campaign).  Clinton is lagging behind Obama's pace in just three states--Nevada, New York, and Ohio--but still leads in these states, with only Nevada and Ohio close at this point.

Clinton's gains in Arizona (+4.4), Georgia (+3.7), North Carolina (+1.7), and Virginia (+2.6) are particularly notable as Virginia was narrow won by Obama in 2008 and 2012, North Carolina went for Obama in 2008 but not in 2012, and Georgia and Arizona have been somewhat out of reach for Democrats.  If Virginia becomes safe for Clinton and the battle for the Electoral College ends up being fought in places like North Carolina, Georgia, and Arizona, it's will be virtually impossible for Trump to carry the day. Utah also stands out as a big gain for Clinton (+17), no doubt in part due to Mitt Romney's vocal opposition to Donald Trump, but I doubt that it will actually be in play on election day.

Of course it is important to note that this analysis is limited to those states for which I could get polling averages from early August in both 2012 and 2016.  However, most of the remaining states are usually less competitive than those discussed above and there is less mystery about where they will end up on election day.

Convention Bumps

As we head into the convention phase of the campaign, I thought it would be useful to review the extent to which the convening party candidate receives a bump in the polls during their convention.  I've covered much of this in previous posts so I'll just briefly review a few key points here.

The figure below summarizes the size of convention bumps from 1964 to 2012.  The convention bump is measured as the percentage point change in the convening party's share of the two-party vote, comparing polls taken between six days and two-weeks prior to the convention with polls taken during the seven days following the convention.  Note that this is a short-term measure of the convention bump and does not say anything about the rate of decay in the weeks following the convention. 

There is a lot going on here but there are a few takeaway points.  

1.  Candidates generally get a bump of some sort.  The size of the bump is highly variable but virtually all candidates leave their convention doing better in the polls than when they went into the convention.

2.  The size of the convention bump does not predict the overall winner very well.  Just ask Presidents Goldwater,  Mondale, Dole, or Gore, all of whom had bigger bumps than their competitors.  One of the reasons for this is that candidates running way behind in the polls have an easier time gaining ground during their conventions.  For instance, in 1964 Barry Goldwater was so far behind in the pre-convention polls (averaging 21% of the vote) that it was easy for him to improve his standing by thirteen points during the convention, though he still never come close to being competitive.  At the same time, Lyndon Johnson went into his convention with 69% of the vote in pre-election polls and left the convention with no bump but still with a substantial lead.

3.  Convention bumps aren't what they used to be (see figure below).   Prior to the 2000 election, convention bumps averaged more than six points, but that has fallen to just over two points from 2000 to 2012.  One potential explanation for this change lies in the scheduling of conventions. The 2008 and 2012 conventions were held in late August and early September and were also held on back-to-back weeks.  The norm in other years had been to hold the conventions in late July or early August and to separate them by two to three weeks.  What is probably most important here is holding the conventions on consecutive weeks, which means that the convention messages end up overlapping and may cancel out each other.  Another potential explanation lies in the increased polarization of the electorate.  It is possible that partisans are so much more committed to their candidates now than they were before and there is a much smaller persuadable electorate that can be influenced by events like the nominating conventions.

So what does all of this mean for the 2016 convention bumps?  One of the the key features of the 2016 conventions is that they follow the recent scheduling trend of back-to-back convention weeks, so this might limit the size of the bumps.  On the other hand, since the conventions are being held in late July rather than late August, there might be more persuadable voters than in 2000, 2008, or 2012.   One other factor that could lead to more substantial convention bumps is that both candidates have problems within their own party and the conventions present them with an opportunity to rally the base in a way that no other campaign event can.  The key for both candidates is to have a smoothly run convention that heals rather than exacerbates existing party wounds and projects a positive message to the rest of the country.