Author: Jim Kane

November 24, 2020

OK the polls were generally off again. They underestimated the Trump vote even after all the changes recommended by independent experts since 2016. In the last post, I showed that Trump voters were not candid about who they were voting for on election day, a form of social desirability effect. “The Return of the Shy Voter?”

As I was analyzing polling data before the Presidential election, I was having some doubts about the polling data. A little voice in my head kept saying “are you sure these numbers are right?”

At the time, I was trying to predict the election outcome in key swing states like Florida. As you may recall, the average of polls for Florida gave Biden a statewide lead of 1.2%. Making a prediction for Florida on that statistic would be like betting on a horse with one fast time but who never won a race. So I called the state for Trump. That’s how scientific it was.

I have always been convinced there had to be a better way to predict an election other than based on polls only. But my previous attempts to find that method have always led to failure.

But that has never deterred me from continuing my gallant quest to discover the magic variable that would solve the election prediction problem. Until now…

I have always been interested in the “favorability rating,” which supposedly measures how voters feel about the candidate. The question is simple: “Do you have a favorable or unfavorable opinion of Donald Trump?”

As you can tell, this question doesn’t measure anything specific like the common Job Approval rating, which measures the incumbent’s performance in office. The favorable rating is considered a popularity measure, but I’m now convinced that it is more than that.

There is a general lack of studies on what the rating means due to a lack of analysis on the subject. It was generally believed that they correlated with job approval ratings, but when Bill Clinton’s affair became public his favorability declined significantly but his job approval rose.

The academic literature has only a few studies about the impact favorability rating except to suggest that it is more than just another way to measure an incumbent’s job approval.

But I found an article on the fivethirtyeight web site published in 2012 entitled “Do Romney’s Favorability Ratings Matter?” https://fivethirtyeight.com/features/do-romneys-favorability-ratings-matter/

This brief study shows that in the late stages of a Presidential campaign “the candidate with the stronger favorable rating in the late stages has won every election since 1980, and there has been an almost-perfect correspondence between the margin of the favorability gap and his margin of victory or defeat in the popular vote.” As important as this statement is, they posted no data to confirm this assumption. But this was my first clue that the favorability rating was more than just a popularity variable.

Some of you know my fondness for job approval ratings and why they matter for the incumbent. But the biggest issue with this rating is it doesn’t apply to non-incumbents who are challenging the President. Consequently, we cannot measure the difference between the two, which could tell us how the voter would likely choose on election day.

But that is not the case with the favorability ratings. It applies to both the incumbent and the challenger, allowing us to make a significant comparison on how voters feel about candidates. At least that is my hypothesis.

But there is one important problem. Most pollsters don’t ask this question at the state level. At least until now. There is a new resource for state by state surveys from a relatively new company called Civiqs.

Their methodology is an opt-in survey panel covering every state in the country (https://civiqs.com/). One of their continuing questions is the favorability rating. Each day the data is uploaded to their public site.

Using this data, I started compiling the favorability rating for Trump and Biden for each state as shown in Table 1.The date for these ratings was October 25, 2020, some nine days prior to the election .

STATE FAVORABILITY RATING	TRUMP	BIDEN
Alabama	62.5	36.4
Alaska	51	40
Arizona	49	49.5
Arkansas	62.6	34.6
Calfornia	33.4	64.6
Colorado	41.9	55.4
Connecticut	40.4	58.1
Delaware	39.8	58.8
Florida	51.2	47.9
Georgia	46	46
Hawaii	25	59
Idaho	63.9	33.1
Illinois	42.3	55.8
Indiana	57.1	41
Iowa	53.2	45
Kansas	56.5	43.3
Kentucky	62.1	36.2
Louisiana	58.5	39.8
Maine	44.2	52.9
Maryland	34.9	63.3
Massachusetts	32.6	65.6
Michigan	47.9	50.6
Minnesota	41	47
Mississippi	59.7	38.8
Missouri	56.9	41.3
Montana	56.9	40.6
Nebraska	58.6	39.2
Nevada	40	45
New Hampshire	45.6	52.8
New Jersey	37	54
New Mexico	43.6	54.2
New York	43	55.7
North Carolina	50.1	48.7
North Dakota	65.5	31.9
Ohio	51	43
Oklahoma	65.4	32.3
Oregon	40.6	57
Pennsylvania	49.1	49.7
Rhode Island	39.1	59.4
South Carolina	55.1	43.4
South Dakota  ?	62	35
Tennessee	60.7	37.4
Texas	52.2	46.4
Utah	58	37.8
Vermont	30.8	66.4
Virginia	44.5	54.1
Washington	38.7	58.8
West Virginia	68.7	29.6
Wisconsin	48.9	49.6
Wyoming	70.4	26.7
AVG. FAVORAABILITY RATING	49.782	47.054

If you look at Florida, you will see that the favorability ratings between Trump and Biden were quite close. There is just 3.3% difference between the two, in Trump’s favor. As we all know now, the Florida contest was close. In fact, when the dust cleared Trump defeated Biden by only 3% (51%-48%) in Florida.

At first, I didn’t see the similarity between the actual vote and the candidate’s favorability rating. When I took a closer look, I noticed that there was a 3.3% difference between Trump and Biden’s favorability ratings taken some 9 days before the election. Just three-tenths of a percent difference than the final results in Florida.

As interesting as that result seemed, I was not ready to say Eureka just yet. But it gave me the incentive I needed to compile by hand all the favorability ratings for all states on that same date, as shown below in Table 2 below.

A negative value indicates that Biden had a negative rating because I subtracted Trump’s percent from Biden’s in every case. For example, Alabama shows the favorability rating difference was -26.1%. That means Trump had 26.1% more favorable ratings than Biden in Alabama. In California, Biden had a positive 31.6%, which indicates he had 31% more favorable ratings than Trump in California.

As a comparison, I have added in the third column the final vote percent difference between Biden and Trump.

As an example, the final election vote for Biden in Maryland was 63.3%. For Trump it was 34.9%. If you subtract 34.9% from 63.3% your get 28.4% (Trump vote % – Biden vote %). The final vote difference on election day was exactly 28.4%!

STATE	BIDEN/TRUMP.FAV.DIFF	FINAL VOTE PERCENT DIFFERENCE
Alabama	-26.1	-26.05
Alaska	-11	-10.1
Arizona	0.5	0.5
Arkansas	-28	-28.5
California	31.2	31.6
Colorado	9	11.25
Connecticut	17.7	18.35
Delaware	19	19
Florida	-3.3	-3.15
Georgia	0	-0.1
Hawaii	34	29.4
Idaho	-30.8	-30.9
Illinois	13.5	12.75
Indiana	-16.1	-18.05
Iowa	-8.2	-8.1
Kansas	-13.2	-13.6
Kentucky	-25.9	-26.45
Louisiana	-18.7	-18.35
Maine	8.7	9.85
Maryland	28.4	28.4
Massachusetts	33	34
Michigan	2.7	2.35
Minnesota	6	6.5
Mississippi	-20.9	-20.45
Missouri	-15.6	-15.8
Montana	-16.3	-15.65
Nebraska	-19.4	-19.7
Nevada	5	5
New Hampshire	7.2	7.6
New Jersey	18.6	20.8
New Mexico	10.6	10.3
New York	12.7	15.35
North Carolina	-1.4	-1.4
North Dakota	-33.6	-33.3
Ohio	-8	-8
Oklahoma	-33.1	-33.05
Oregon	16.4	16.7
Pennsylvania	0.6	0.65
Rhode Island	20.3	20.15
South Carolina	-11.7	-11.85
South Dakota	-27	-26.5
Tennessee	-23.3	-25.15
Texas	-5.8	-5.9
Utah	-19	-19.5
Vermont	35.6	34.3
Virginia	9.6	9.3
Washington	20.1	21.05
West Virginia	-39.1	-39.05
Wisconsin	0.7	0.35
Wyoming	-43.7	-43.35

Out of the 50 states, the largest error rate between Trump and Biden’s final vote difference and their favorability difference was in Hawaii, with 4.6% difference. For all 50 states, the average error difference was only .04%.

In political science, we use regression models to determine the relationship between variables which also produces a scatterplot of the results. In the Chart below, this scatterplot shows how tight the “fit” is for the models estimates.

Each small circle represents a state and the red line the linear relationship between the Biden / Trump favorability difference and the final vote percent difference. From a statistical viewpoint, the relationship between the two measures is almost perfect (.994 R square).

In other words, we can now predict the final vote percent difference between two candidates by using only the favorability ratings prior to the election. More importantly, the error rate is only .04%.

Does this method apply to the national vote? The national favorability rating for Biden on October 25th was 45% and for Trump 42%, a 3% difference. The final national vote for Biden was 51.1% and for Trump 47.2%, a 3.9% difference. The final vote difference was 3.0%, an error of nine-tenths of one percent.

So why has no one discovered this by now? There a couple of possible reasons. First, there is very little research on the favorability ratings and what they mean to voters. Secondly, until recently, no polling firms have consistently used these ratings at the state level. Finally, with the exception of Nate Silver’s fivethiryeight review, no one else has seen any relationship between the ratings and the final election results.

The bad news is that Civiqs data does not go back to 2016, so we can’t compare the ratings to the actual vote in that Presidential election.

Who knows, in the future we might be able to cancel the election altogether and just rely on this system instead. I’ll have to check with my pocket Constitution to see if that is legal. Be safe…

November 16, 2020

I’m sure you have heard or read that the Presidential Election polls screwed up again. Initially I was skeptical, of course, since 2016 almost all polling firms had made significant changes to their methodologies such as weighting voter education.

But after all the studies and soul searching, whatever changes that were made did not cure most if not all the errors, at least to Donald Trump’s share of the vote.

Using only swing states and not the national vote, I have compared the polling averages from surveys taken from polls conducted within the final 30 days prior to the election.

The states I analyzed are Florida, Georgia, Pennsylvania, Wisconsin, North Carolina, Arizona, Michigan and Ohio. No polls were excluded because of political bias or voter interview methods (live-interviews, online, or IVR, etc.).

I computed the absolute error between the polling averages and the actual state election results for both Biden and Trump. For example, if the final state result had Trump with 48% and the polling average at 46%, that would mean an error of 2%.

In the table below are the Trump and Biden absolute error rates for each state, starting with North Carolina, which had a Trump polling error of 2.3% and for Biden, 1.1%.

If you go down the list, all state polls except Florida, had Trump’s polling percent with greater error than for Biden. In all eight states, Trump’s average absolute error was 2.93% and Biden’s .062. All the state differences were statistically significant (T-Test).

	TRUMP	BIDEN
STATE	ERROR	ERROR
N.CAROLINA	2.3	1.1
OHIO	6.2	-1.05
PENNSYLVANIA	1.6	1.1
WISCONSIN	5.1	-1.4
GEORGIA	2.3	1.1
MICHIGAN	2.9	0.1
ARIZONA	2.6	0.4
FLOIIDA	0.4	3.6
AVG. % ERROR	2.93	0.062

The average error for all of these states for Trump’s share of the poll percent’s was a significant 2.9%. But for Biden it was only .062%. In other words, the polls consistently underestimated Trump’s vote by nearly three percent.

Random Error vs. Systematic Error

There is some error in every poll and there is nothing we can do to prevent it. Of course you can increase the sample size, but that can only reduce the error. Pollsters expect this, but they often pretend that polls properly conducted represent the future outcome. Polls are estimates and not absolute predictions.

And then there is systematic error. Before the survey begins, pollsters try to predict the turnout. Who votes and those who do not can alter that polling estimate regardless of statistical error. Overestimating or underestimating one group can have a significant effect on the final results. That is systematic error.

When I saw the Trump vote errors, I felt like Yogi Berra when he said “It’s like déjà vu all over again.”

So how did lightening strike twice in two elections. After 2016, the best minds in survey research came up with few potential errors:

1. Undecided voters broke for Mr. Trump in the final days of the race. But that didn’t happen this year according to exit polls and independent surveys.

2. Turnout among Mr. Trump’s supporters was somewhat higher than expected. No, that did not happen this time. Pollsters expected high turnout from both Biden and Trump voters.

3. Trump’s support in the decisive Rust Belt region, in part because those surveys did not adjust for the education. Not this time: Every major pollster adjusted for education in this election.

4. Some type of “shy Trump” voter bias was dismissed as unlikely.

What stands out to me is how the Biden poll estimate was right on the money, with an error rate of only .062%, which is about as perfect as it gets in surveys.

But the Trump voter error was almost 3%. This is undoubtedly a systematic error and not random. And this is the second Presidential election this has occurred.

The obvious answer to the failure of polls for the last two election cycles are with the responses of Trump voters. The average Biden poll percent of the vote was on the mark. All the error came from the responses of Trump supporters.

So let’s narrow down the search for the systematic error as to why Trump voters were underestimated when every other possibility has been eliminated.

We have eliminated previous possibilities, so that suggests that either Trump voters refused a polling interview, didn’t answer the phone or misled as to their real preference.

In the later case, this is a type social desirability effect called self-deceptive enhancement (SDE). SDE response bias is any systematic tendency to answer questionnaire items on some basis that interferes with accurate self-reports.

Self-deceptive enhancement bias is common in surveys on sensitive issues such as job performance surveys and sexual behavior. Pollsters have known this for decades. The cure for this bias in many cases is the application of computerized surveys.

But the issue of who you are voting for has never been considered sensitive enough to cause significant errors.

If self-deceptive effects are occurring, then we should see more accurate results from firms using Interactive Voice Recording (IVR) and online surveys, where the voter expresses their choice to a machine and not to a live person. This would be similar to a computerized survey when an honest/accurate response is important.

In my insatiable quest for truth in polling, I compiled a list of polling firms by their interview technique: live calls, IVR (robocalls) and online surveys listed on 538’s website.

Then I compared their interview method with the their last months Trump survey results and compared it to the final vote percent for Trump. (31 polling firms met this criteria). If there is no bias, we should see similar results from all three methods when compared to the actual vote.

My hypothesis is that the IVR and online polls were significantly (<.001 level) different than live caller polls and more accurate in Trump’s final vote. This would mean that Trump voters were not truthful in their responses to the live interviewers as compared to non-live caller interviews (IVR/online). In other words, the Trump shy-voter lives!

The results to this experiment were conclusive. As the chart below graphically demonstrates, live-interviewer polls have a significantly larger error rate for Trump supporters than either IVR or IVR/online polls, with an average error rate of 5.62% (red bar).

The survey method that was the most accurate on the Trump responses was the IVR/Online surveys, with an average error rate of only 1.6%. Next accurate was the IVR poll, with an error rate of 2.85%. Bringing up the rear, of course, are the respected and dominate forms of polling – the live-caller interviewer surveys with a 5.62% error rate!

The survey firms in this last category are the country’s most respected and considered the most accurate pollsters in the country. You know who I’m talking about.

I’m not going to criticize these major firms since I quite frankly didn’t believe that social desirability effects were causing these survey errors either until I looked at the data.

But in hindsight, I should have considered it. After all, Donald Trump considers the Proud Boys as a socially acceptable group. For many Republican voters, this is probably not an acceptable opinion (along with many others he has espoused). Consequently, when talking to a live-interviewer, they made a small fib, even though they planned to vote for Trump.

This insight, if used in Presidential election surveys where one of the candidates has opinions that are outside the mainstream, such as Donald Trump running again in 2024, should cure this problem.

Let me hear from you on what you think. I could use your comments. Be safe…

November 2, 2020

Ok, tomorrow is Election Day (thank God). And with caution thrown to the wind, I will make my key state predictions. Using the latest large sample surveys that are available, and the early partisan votes to confirm my intuition.

First, the basics: All surveys have must have at least 700 or more interviews. No surveys are excluded for their political bias, as long as their methodology is acceptable.

Florida: The Swinging Shine State

DATE	BIDEN	TRUMP	MOE
10/29 – 10/31	48	47	3.5
10/29 – 10/30	51	45	3.3
10/29 – 10/30	49	48	1.9
10/27 – 10/31	47	44	3.2
10/26 – 10/29	50	47	—
10/25 – 10/28	47	50	2.9
10/24 – 10/29	48	50	4
AVG.	48.6	47.3	3.1

These two charts display the current averages between Biden and Trump. With only a day to go, Biden in the last seven polls has 48.6% versus Trump’s 47.3%. That’s only a 1.3% difference and far less a difference than the average margin of error of +/-3.1% (MOE).

But when you look at the early partisan voting turnout, Democrats have only a 108,000 vote lead over registered Republicans. Now I understand that you can’t know how these partisan voters voted, but most will vote along party lines.

The problem is that a 108,000 may not be enough Democratic voters to offset the lead the Republicans can generate on election day. This is the day that Republicans come out of the woodwork to vote. And many Democratic leads have failed to make the finish line after all the votes are cast.

That said, the large NPA turnout may offset this advantage. As I have posted recently, NPA voters have a 58% probability to vote Democratic. This could male the difference for a Biden victory.

It’s obvious that from a statistical point of view, Florida is a tossup and we could wait several days, if not weeks to find out who won. My guess: Donald Trump. When it’s close, the Republicans always seem to find a way to win in Florida.

Doctor Politics Florida Call: Trump

Arizona

DATE	BIDEN %	TRUMP %	MOE
10/29 – 10/31	48	46	3.6
10/27 – 10/29	45	49	3.5
10/26 – 10/30	49	43	3
10/23 – 10/30	50	46	3
10/25 – 10/28	46	49	4.1
AVERAGE	47.6	46.6	3.44

Well if you thought Florida is close, Arizona makes Florida look like a landslide. The average of the last five polls shows Biden with a one point lead, with a margin of error of +/-3.44%. If you look up the word “tossup” in the dictionary you would see a map of Arizona.

And early voting gives the Democrats a lead of only 43,055 votes. This gives Trump a pretty good shot on winning this state.

DR. POLITICS NORTH CAROLINA CALL: TRUMP

GEORGIA POLLS:

	BIDEN	TRUMP	MOE
10/29 – 10/31	48	49	3.5
10/28 – 10/28	47	48	3.6
10/27 – 10/28	48	46	3.8
10/23 – 10/27	50	46	4.4
AVERAGE	48.25	47.25	3.825

Georgia hasn’t voted for a single Democratic candidate since 1996. But as you can see, the last four larger sample polls, the difference is only one percent.

Georgia does not require a party registration, so we don’t know the partisan early/mail vote party differences. Atlanta has the largest proportion of Democratic voters. So far 780,000 Atlanta votes have been cast, out of a total vote off four million votes cast so far, or about 20% of the vote.

I’m sticking to tradition: Dr. Politics Calls Georgia for Trump

THE BLUE WALL RISES

The Keystone State: Pennsylvania

DATE	BIDEN	TRUMP	MOE
10/31 – 11/1	50	47	3.5
10/30 – 10/31	46	48	2.9
10/29 – 11/1	51	46	4.4
10/27 – 11/1	51	45	4.3
10/27 – 10/31	49	43	2.4
	49.4	45.8	3.5

Next to Florida, this is a state that really matters if you are a Trump fan. If he loses this state on Tuesday, Donald Trump can go ahead and rent that moving van. (He will, of course, contest this and several other state totals, so he can probably wait a few more weeks.)

Biden is leading in every recent poll except
Trafalgar, which is the pollster who called Trump winning in 2016. They have Trump up by 2 points. The average Biden lead among all surveys is 3.6%, just above the margin of error.

As far as the early vote, Democrats have over a one million mail vote lead over Republicans going into election day.

Dr. Politics Call: Biden

Michigan

DATE	BIDEN	TRUMP	MOE
10/30 – 10/31	46	48	2.9
10/29 – 10/30	52	45	3.4
10/27 – 11/1	52	42	4
10/29 – 10/29	52	45	3.4
	50.5	45	3.425

In the last four Michigan surveys, Biden leads in all but one poll. It is, of course, the Trafalgar survey with Trump leading by 2%. On average, Biden leads by 5.5%, above the average margin of error of 3.4%.

Wisconsin: The Blue Wall Holds

Excluding Trump’s narrow win in 2016, no Republican presidential candidate has won this state since 1988.

DATE	BIDEN	TRUMP	MOE
10/29 – 10/30	53	45	3.5
10/27 – 11/1	53	43	4.2
10/26 – 10/30	52	41	3.2
10/24 – 10/25	48	47	2.9
10/23 – 10/30	52	44	3.9
10/21 – 10/25	48	43	4.4
AVERAGE	51	43.8	3.7

And as we enter election day, it looks like this trend will continue. Biden now has a 7.2% lead. The average margin of error is +/-3.7%, absent 2016 polling errors, this looks Biden has this state’s 10 electoral votes.

IMPORTANT: Biden can win the electoral college vote without Florida, Georgia, North Carolina, Ohio, Texas, Arizona and Nevada, if he carries Pennsylvania, Wisconsin, Minnesota, and Michigan. (And the other states Hilary won easily in 2016).

This would give Biden 280 Electoral Votes and the keys to the White House.

If you are concerned a 2016 poll redux, I don’t seeing this happening in any significant way. All the pollsters I have talked to have resolved many of the weighting issues that caused most the errors. More importantly, we have far more state-level surveys than we did in 2016 making the averages more reliable.

As for watching the election results on TV, make sure you have plenty of alcohol on hand. It’s going to be a long night… be safe.

October 28, 2020

Florida has a total of 3,783,286 No Party Affiliation voters (NPA). Over a million have already voted in the Presidential Election. If you are a Democrat or a Republican you have a high probability of voting for the candidate representing that Party. But what about the NPA’s? Most folks call them independents, but as I have posted previously, most independents tend to lean toward one party. http://thepoliticsdr.com/wp-admin/post.php?post=1735&action=edit

Political pundits on TV show polling results comparing men vs woman, young vs old, whites vs blacks will vote. But you never see surveys that compare NPA’s vs Republicans or Democrats. NPA’s do vote in presidential elections, but are largely absent in local elections. At this point, 1,395,712 NPA’s have already cast their vote. That’s 20% of the early vote.

But is there a way to estimate how the NPA registrants might vote based on the 2016 election data to see how they might vote on Tuesday? The answer is yes, but it is a little complicated. So bear with me and I will simplify as best I can.

Using election data from Florida’s 2016 Presidential Election, in particular voter registration data at the county level, I designed a model that estimates the percentage of NPA voters who voted for Clinton.

The statistical method is called multiple regression, which I have used in previous posts. In political science this is the most used model in academic studies.

A simple explanation for this statistical method is that it allows for the control of other influential variables, thus isolating the independent impact of the variable of interest, which in this case is the county level registration of NPA voters on Hillary Clinton’s total vote.

The table below shows the models explanatory statistics. Focus on the R square statistic which is .998. This means that nearly 100 percent of the variance in the dependent variable (Clinton’s vote) is explained by the model. That’s as high as it gets.

Model Summary
Model		R Square	Adjusted R Square	Std. Error of the Estimate
		.998	.997	.997

In order to isolate other variables that could affect her vote, we need to control for the county level registration of Democrats and Republicans which would, of course, have a significant effect on her vote.

Coefficients
Model	Unstandardized Coefficients
	Std. Error	Beta .SIG
	(Constant)	-879.101 .490
NPA.REG	.578	.075 .000
DEMOCRAT	.674	.040 .000
REPUB	-.203	.034 .000

Notice how the Republican registration unstandardized coefficients variable (REPUB) has a negative sign, meaning it reduces her votes which we would expect. The Democratic variable has positive sign, which indicates it has a positive effect on her vote. And the NPA registration also has a positive sign as well.

To see how the model actually performs we need to see how it performs on the 2016 election, which is shown in the Chart 1 below. It displays the models scatterplot of the NPA registration with Clinton votes based on the model’s estimate.

In the Chart, the small circles are the 67 Florida counties. As the number of NPA registrations increase, so does the total county votes for Clinton as well. The straight line reflects how linear the relationship is between NPA registration on her vote while controlling for Democratic and Republican’s county registration.

What does this mean? Well it means we can now estimate

Clinton’s likely vote from NPA voters alone. If you look back at Table 2, labeled “Coefficients,” you will see a column labeled Unstandardized Coefficients. Next to the NPA REG, is the .578 coefficient. What this tells us is that a one unit change in the dependent variable (Clinton vote) changes if we increase the independent variable by one unit (.578) keeping other independent variables constant.

OK, let’s use examples of how this affects Clinton’s vote. For every increase of NPA registration by .578 increases Clinton’s vote by one. In plain terms, an increase of 100 NPA registrations increased her vote by 58.

Now let’s apply this estimate to Biden’s vote. Currently, 1,395,712 NPA’s have already voted and using our estimate that would mean that 806, 721 voted for Biden. In other words, he is likely getting 58% of the NPA vote.

Let’s estimate that 75% of NPA voters participate in this election. That means that 2,837,465 total NPA votes, which would add 1,645,729 votes to the Biden total column. That is 16% of the total vote if we have a 75% turnout.

Now for the caveats. First, the model is an estimate based on the a previous election. Secondly, the model should be considered a broad estimate. Saying that, what this analysis does show is that Democratic candidates benefit from higher NPA turnout.

Personally, I don’t know if either the Florida Democratic Party or the Biden campaign had a get out the vote program for NPA voters. If they haven’t they should add one to swinging state in the future. Be safe and vote…

October 27, 2020

This a quick update on seven key battleground states. In the chart and table below, is a one month comparison of Joe Biden’s average poll lead in these key states.

POLL DATE	23-Sep	25-Oct
STATE	BIDEN LEAD %	BIDEN LEAD %
FLORIDA	1.5	-0.4
WISCONSIN	6.9	5.5
MICHIGAN	6.5	9.4
PENN	3.8	3.8
MINNOSOTA	10.2	6
ARIZONA	4.4	2.4
N. CAROLINA	0.7	0.8
AVERAGE	4.86	3.9

The last two columns are the end dates of each series of surveys. First series is Biden’s lead in seven battleground states completed on September 23. The second series on October 25, a little over one month apart. The chart shows a graphical representation of each state’s changes during this period.

In all seven states, Biden lost on average one percent. The only state he increased his lead was Michigan, where he improved his lead by nearly 3%. And he is still ahead of Trump in the key state of Pennsylvania, at 3%.

He has fallen slightly behind Trump in Florida. In general, however, Biden is holding his own with only one week to go. But its clear that Florida, Pennsylvania, Arizona and North Carolina are still very close and still in play. That’s 75 electoral votes.

Make sure you get a nap on November 3rd, cause it looks like its going to be a long night. Be safe …

October 26, 2020

As expected, early voting is benefiting the Democrats. Data collected by the Elections Project (United States Elections Project) from the Florida Division of Elections shows that Democratic voters as of October 25, have a 6.4% lead over Republican voters, as shown below.

Total Voted by Party Registration	OCT. 25
Party	Count	Percent
Democrats	2,440,470	42.8
Republicans	2,076,621	36.4
Minor	69,798	1.2
No Party Affiliation	1,119,404	19.6
TOTAL	5,706,293	100

As of this date, Democrats have a 363, 849 vote advantage over the Republicans.

Most of this early Democratic advantage comes from Mail-in ballots. In Table 2 below, you will find the Democrats have returned 594,110 more ballots than Florida Republicans and have a return rate of 65% compared to the Republicans 61.8% rate.

Mail Ballots Returned by Party Registration	25-Oct
Party	Returned Ballots	Freq. Distribution	Requested Ballots	Return Rate
Democrats	1,744,527	47.1	2,671,740	65.3
Republicans	1,150,417	31	1,862,285	61.8
Minor	45,116	1.2	82,195	54.9
No Party Affiliation	765,729	20.7	1,353,719	56.6
TOTAL	3,705,789	100	5,969,939	62.1

Where the Republicans excel is in the in-person voting. As displayed in Table 3 below, the Republicans lead by 11.5% in this category.

In-Person Votes by Party Registration
Party	Count	Percent
Democrats	695,943.00	34.80
Republicans	926,204.00	46.30
Minor	24,682.00	1.20
No Party Affiliation	353,675.00	17.70
TOTAL	2,000,504.00	100.00

At this point, Republicans have outvoted in-person the Democrats by 230,261 ballots. And on Election Day, this is where the Republicans will close the gap.

With the Democrats significant lead in Mail-in ballots, the Republicans need to make up a deficit of 363, 849 votes on November 3rd. But don’t underestimate the Election Day Republican vote. This is when they start showing up at the polls. This lead will evaporate quickly.

In 2016, there were 9,122,861 total ballots cast which equated to a 75% turnout of registered voters.

2016	VOTES CAST
TRUMP	4,617,886
CLINTON	4,504,975
TOTAL	9,122,861

If the 75% turnout rate holds, that means there are still at least 5,125,108 votes still to be cast. That’s still a lot of votes and only a week to cast them. If you haven’t voted already, make sure you are part of that 5 million votes. Be safe…

October 23, 2020

The impact of the incumbent’s Job Approval rating on his reelection nationally is well established. Since 1948, the beginning of modern polling, only one incumbent president, Harry Truman, has won reelection with an job approval rating less than 48%.

Below is a table showing all incumbent presidents since that election with their job approval ratings in June and just prior to the election.

PRESIDENT	YEAR	June of reelection year	Final measure before election	Won reelection
		% Approve	% Approve
Obama	2012	46	52	Yes
G.W. Bush	2004	49	48TRTR	Yes
Clinton	1996	55	54	Yes
G.H.W. Bush	1992	37	34	No
Reagan	1984	54	58	Yes
Carter	1980	32	37	No
Ford	1976	45	n/a	No
Nixon	1972	59	n/a	Yes
Johnson	1964	74	n/a	Yes
Eisenhower	1956	72	68	Yes
Truman	1948	40	n/a	Yes
GALLUP SURVEYS

As the table shows, eight of eleven of incumbents won reelection during this period, and no President won with a job approval of less than 48% (Truman). And the average approval rating is 51.2%.

Political scientists believe this one metric more than any other, determines an incumbents reelection chances. This makes sense if you think of a reelection as determining whether an employee, the chief executive of the nation, should be retained. If he has done a good job, you keep him on for another four years. If he hasn’t, you send him packing even if you like him.

The job approval poll question was created by George Gallup for that one reason. If you’re wondering, as of this date Trump’s national average Job Approval is at 44.5%. And his average rating since he was elected stands at 40%, the lowest average since the question has been asked.

It is possible that Trump could break this record and win the November 3rd election. But that isn’t my interest here. I’m interested if the job rating can be applied at the state level.

In other words, can Donald Trump win Florida, for example, with a statewide job approval of less than 48%, or does this metric only apply to national elections?

In my quest, I scoured Google Scholar for any academic articles that would answer this question and came up empty handed. I could not find a single article that addressed this question at all.

The reason for this dearth of scholarship on the subject should have been obvious to me from the start. State polls don’t ask this question, except in rare cases.

But recently, Civiqs has developed an large opt-in online panel that conducts daily surveys of randomly selected (list based) registered voters from its large panel from all 50 states.

A regular question asked each day is the standard job approval rating. And from their database, I have now found job approval ratings for all 50 states as of this date. I want to stress that this exercise is experimental and the likelihood that a state approval rating could predict the winner is unlikely. (That’s my null hypothesis.)

As a baseline, I am using a 2017 Gallup state by state survey using the standard job approval rating. At this point I am only interested in these swing states: Arizona, Florida, Georgia, Iowa, Michigan, North Carolina, Ohio, Pennsylvania, Texas, and Wisconsin. But I may add the other 40 as we get closer to the election.

In Table 2 below, lists the 2017 approval rating and the current October 22, 2020 rating in each of these states. In the fourth column, is the percent difference between the two surveys, which are all positive. The last column is Trump’s current average poll percent lead or deficit in each state.

For example, in Arizona, Trump had a job approval rating of 41% in 2017 and a current rating of 44%, which is 3% higher than in 2017. In the latest average of polls, Trump is behind Biden by 3.2%.

	TRUMP JOB APPROVAL 2017-2020
STATE	2017	2020/CURRENT	DIFF%	POLL%
Alabama	53
Alaska	50
Arizona	41	44	3	-3.2
Arkansas	50
California	29
Colorado	37
Connecticut	31
Delaware	36
Florida	41	45	4	-1.5
Georgia	41	45	4	-1.2
Hawaii	29
Idaho	53
Illinois	33
Indiana	44
Iowa	43	47	4	-0.6
Kansas	48
Kentucky	51
Louisiana	49
Maine	42
Maryland	30
Massachusetts	27
Michigan	40	41	1	-7.8
Minnesota	37
Mississippi	48
Missouri	47
Montana	52
Nation	38
Nebraska	49
Nevada	42
New Hampshire	42
New Jersey	34
New Mexico	35
New York	30
North Carolina	40	44	4	-1.5
North Dakota	57
Ohio	45	46	1	0.6
Oklahoma	53
Oregon	36
Pennsylvania	42	42	0	-5.1
Rhode Island	32
South Carolina	48
South Dakota	54
Tennessee	50
Texas	39	48	9	4
Utah	48
Vermont	26
Virginia	37
Washington	34
West Virginia	61
Wisconsin	41	43	2	-4.6
Wyoming	57

As the table shows, Trump’s statewide approval rating has increased since the 2017, but not by a lot. The biggest gain is Texas, where his approval rating improved by 9 points and is now at 48%, which at the national level, would theoretically reelect him. His current Texas lead over Biden is now 4%. These two metrics would suggest Texas will go to Trump.

His next highest approval rating is Iowa, at 47%. But Biden now has a narrow lead of just 0.6%. This state is too close to call.

At an approval rating of 45% are both Georgia and Florida. In Georgia, Biden has a 1.2% lead over Trump. Although very close, this state could easily go Democratic this year, if the job approval rating has the same impact it does at the national level.

Now we have the swingiest state in the nation, Florida. Trump’s approval rating is now 45% as well. And Biden has a 1.5% average lead in the state. My gut feeling is that Trump will win the Sunshine State, but at this point the tea leaves give a slight edge to Biden.

In Arizona, Trump’s job approval rating is now 44% and Biden with a 3.2% lead. For now the sate is leaning toward Biden.

Following Arizona we have another close contest in North Carolina, where Trump’s approval rating is 44%, and Biden has a slim poll lead of only 1.5%. Giving weight to the job approval rating, this state looks like it could end up in the Biden column.

And finally, we have the mid-west trifecta of Pennsylvania, Michigan and Wisconsin. All three states have Trump’s approval rating at 43% or less. The average poll lead for Biden in Pennsylvania is now 5.1%, in Michigan it’s 7.8% and Wisconsin at 4.6%. At this point with only 11 days left, all three should deposit their electoral college votes in the Biden column.

I again want to emphasize that this is an experiment to see if the Presidential Job Approval rating has the same predicted capability it does at the national level. Closer to November 3rd, I will update this table.

Feel free to use these tables to place bets with your friends and neighbors. My fee is only 5% on your winnings. I don’t know about you, but I’m excited. Be safe…

October 20, 2020

I have seen some TV reporting that Republicans are outperforming Democrats in new registrations. This stimulated my interest to see if this was true in Florida and if so, what impact it might have on the Presidential Election occurring in two weeks.

I have assembled Florida registration data from 1972 through 2020 as of this years’ book closing, and compared the differences between the two parties. Let’s start with Florida’s total registration over that 48 year period, as shown in Chart 1 below.

The chart shows how Florida’s total registration has increased exponentially over the past 48 years reflecting the state population growth. In 1972, total registration was 3,487,458 and in 2020 is 14,441,869, a 314% increase.

But the real question is what has been the Democrats advantage over this same period and is it increasing or decreasing over this same period. Chart 2 graphically shows how the Democrats registration advantage has declined significantly since 1972.

A little history can explain some of this change. Florida was a traditional Southern state for most of its existence. Southern states were solidly Democratic after the Civil War. Beginning in the 1970’s, Florida’s population changed as new residents arrived from the Midwest and Northeast changing both the population and the culture of Florida’s Old South traditions. At the same time, many old South Democrats changed their registration to Republican.

As the chart shows, there was a sharp decline in the Democratic registration advantage in the 1980’s, finally leveling off around 2002. But in 2008, when Barack Obama was the Democratic nominee, there as new surge in new Democratic registrants.

But when his second Presidential term came to an end, the decline in the Democratic advantage in registration started again, reaching its current level of 134,242 more Democrats than Republicans, shrinking some 560,000 since 2008.

The obvious question is whether the Democratic registration advantage affects which party’s candidate wins? The common sense assumption is that an increase in a party’s registration would increase the likelihood of victory. In my experience, I have found that common sense has little to do with politics.

If an increase in the Democratic advantage increases it should also increase the Democratic election wins. Consequently, the decline in that advantage should increase the likelihood of a Republican win. That’s common sense.

But that is not the case. As you can see in Table 1 below, from 1972 through 2016

YEAR	PARTY	DEM ADV
2016	REP	330428
2012	DEM	558272
2008	DEM	694147
2004	REP	367884
2000	REP	379086
1996	DEM	430773
1992	DEM	645597
1988	REP	903671
1984	REP	1417136
1980	REP	1657782
1976	DEM	1611972
1972	REP	1419605
	AVG	868029
	TABLE 1

In this period, Republicans won seven of the 12 Presidential Florida elections, when the average Democratic registration advantage was 868,029 voters. The year when the Democratic advantage was at its highest (1980), the Republicans won the election.

I also performed a correlation analysis using both the number of votes and number of party victories, and none were statistically significant.

I have to say I expected to see some relationship between registration and the election outcome. In writing this post, I called an old friend who has been directly involved in many campaigns and asked what he thought. His immediate response explained it all: “Candidates still matter.”

Be safe…

October 19, 2020

Conventional wisdom has a higher turnout benefits the Democratic Party. This conclusion is mostly based on the majority party having a large population of lower social economic status (SES) voters, thus decreasing their participation rate on election day.

But in a Presidential Election, the turnout rate is always higher and conceptionally, should benefit the Democratic candidate, everything else being equal.

Another problem with this question is what is a non-voter? For some studies this means someone who is eligible to vote, but doesn’t. That’s in contrast to those registered to vote but doesn’t vote.

For purposes of keeping a level research field, I prefer measuring turnout based on those already registered to vote and not the voter eligible population.

And since we don’t know what the turn-out will be on November 3, I will use data from the 2016 Florida Presidential Election to create a model to estimate what affect it could have on Biden’s percent of the two party vote, while controlling for other variables.

In political science, this requires the application of a statistical method called multiple regression. This method allows for determining the specific effect of one variable while controlling for the effects of other variables in the equation. Specifically, how a one percent increase in turnout affects, if at all, Biden’s percent of the total vote.

This sounds technical and it is, but I will attempt to plainly explain this process as simply as I can. I have taught this to my graduate students for many years and occasionally, some students understand it. But don’t worry, the bottom line: does the model work as an estimate of turnout.

I’m using the county level data from all 67 counties in the 2016 Florida Presidential election. This data file contains approximately 45 different variables that might have some impact on turnout. You cannot go out and find all of this data in one place, because I have personally collected this data from many many different sources.*

Chart 1 below, shows the model’s accuracy estimate. The most important number is the R Square value, which is .901. This means that the model explains 90% of variance in the prediction value. In the social sciences, this is about the highest level you can get from a regression model.

		BASIC STATISTICS
R	R Square	Adjusted R Square	Std. Error of the Estimate	Durbin-Watson
.949	.901	.894	4.40508	1.898

Table 2 below, is the most important table of the group. It includes the statistical significance of each of the variables in the equation. The significance level (sometimes called the P-Value) tells us if the variable has a statistical effect on the dependent variable, which in this case is Clinton’s turnout percent. Remember, I’m using the Clinton data to estimate the affect on Biden’s percent of the vote.

Coefficients Table
Model	Unstandardized Coefficients	Standardized Coefficients	t	Significance
B	Std. Error	Beta
(Constant)	133.006	9.201
REPUB.PER	-1.934	.103	-1.466	.000
DEM.PERCENT	-1.199	.120	-.925	.000
TURN.PER.2016	.308	.137	.102	.029
BLACK.PERCENT	.457	.083	.319	.000

Acceptable significance levels are from .000 to .05. All but the turnout percent in 2016 is at the highest level. But even the 2016 turnout percent is at the acceptable level of .05. I have eliminated all other variables that don’t closely reach these levels.

In Chart 1, I show a scatterplot of the model’s estimate on the affect on Hilary Clinton’s turnout percent based on the models estimates.

As you can see, the model shows a strong linear relationship on her turnout percent in every Florida County (small circles) in 2016.

Now we can use this models unstandardized coefficient (.308) located in Table 2 of the 2016 turnout percent (TURN.PER.2016).

 An unstandardized coefficient represents the amount of change in the dependent variable Y (percent of Biden’s vote) due to a change of 1 percent of independent variable (turnout) in 2020, while controlling for the effects of the other variables in equation.

Specifically, I have controlled for the Republican registration, Democratic registration and the Black registration in each of the 67 counites, so we can see the singular effect of voter turnout alone.

While controlling these variables, if the turnout in Florida increases by 5% over the 2016 turnout (80%), Biden’s share of the two party vote increases by 1.5% percent (5 x .308). If, however, Florida hits the record turnout of 83% in 1992, Biden’s percent of the vote increases by 2.5% (8 x .308).

The increase doesn’t seem major, but we have to remember Florida’s habit of elections that are traditionally razer thin. In 2016, Donald Trump only won Florida by 1.2 %. Theoretically, a 4.5% increase in the turnout would have given Hillary the Florida win. But Trump would still have won the national election with a electoral count of 275 votes even while still losing Florida.

As a caveat, this an estimate derived from one specific campaign. But the model does confirm that in Florida, increasing the total turnout of registered voters does benefit the Democratic candidate. Which confirms the assumption of conventual wisdom. But the effect is small… Be safe…

* For those who want a copy of the data file, you will need SPSS software to use it. Just put a note in the comment section with your email.

October 15, 2020

Many political pundits are predicting a record turnout in the Presidential Election on November 3, including Florida. I don’t disagree with these opinions, based on early statistics of mail voting and early voting in states around the country. Early voting begins in Florida next Monday and I expect to see lines wrapped around early voting sites.

All these opinions of record voting made me wonder, what would be a record turnout in the Sunshine State?

In this exercise, I’m using data from the Florida Department of State’s data site. These turnout figures are based on the percent of registered voters and not either voting age voters (VAP) or eligible voters (VEP). This gives us equivalent comparisons back to 1972.

In Chart 1 and Table 1 below, you will find that by far the largest turnout occurred in 1992, with a turnout of 83%.

YEAR	PERCENT	PARTY WON
1972	74	R
1976	77	D
1980	77	R
1984	75	R
1988	73	R
1992	83	R
1996	67	D
2000	70	R
2004	74	R
2008	75	D
2012	72	D
2016	75	R
AVG	74.3

If you were wondering what caused this difference, 1992 was the year Bill Clinton, George H.W. Bush and Ross Perot faced off against each other.

For those who remember this election, it ways a fascinating combination of three completely different candidates for President. And it was the election where Ross Perot gave his memorable statement on NAFTA during a debate: “there will be a giant sucking sound going south.”

In the following election cycle, 1996, the lowest turnout occurred at 67%, a 16 point drop from 1992. This election again had three candidates, Bill Clinton, Bob Dole and Ross Perot and it was boring and predictable.

Perot had lost his luster and captured 9.2% of the vote, and Dole with 42.3%. Incumbent Bill Clinton managed 48% and won both Florida and the national vote.

Below is a normal probability distribution of all turnout rates since 1972. This is also called a “bell curve.”

Since this is a normal distribution, we would expect 68% of all turnout rates to be within one standard deviation of the mean (74.3%), which is 3.9%. Simply put, the odds suggest that there is a good chance that the next turnout rate will be between 78% and 70%.

I know what you are thinking, “what good is that prediction?” Well that’s the problem with statistics, it only tells us the probability of an occurrence and not an exact prediction. Otherwise, I would be now living on my 100 foot yacht on the French Riviera.

There is no doubt that this election has created a huge amount of interest for supporters of both candidates in Florida. Early anecdotal evidence suggests right now the momentum favors the Democrats.

But will it beat the Florida record of 83%, nearly 10 points above the mean? That’s a big jump from any other previous year. With that said, there has never been an election quite like this in all of Florida’s history either.

I don’t know about you but I’m putting my money on a record turnout at 85%. Don’t worry if I lose, I never bet more than $10. Be safe…