Russell S. Sobel
West Virginia University
email: rsobel2@wvu.edu
homepage: http://www.be.wvu.edu/divecon/econ/sobel/
and
Robert A. Lawson
Capital University
homepage: http://capital2.capital.edu/faculty/rlawson/
November 15, 2000
Many strange things may have happened in the state of Florida during the presidential election of 2000. As never before, the internet has allowed a rapid dissemination of research on these events. Within one week after the election, at least 20 different scholarly analyses of the “butterfly” ballot in Palm Beach County issue have been published on the internet (see both Greg Adams’s web page at Carnegie Mellon University http://madison.hss.cmu.edu/ and Jonathan O'Keeffe's web page http://www.bestbookmarks.com/election/#links). At the time of this writing only one, however, has even barely explored the issue of “the central time zone hypothesis,” which potentially could account for enough votes to alter the outcome in the state of Florida (see http://www.stanford.edu/~jromley/election.html which finds a small pro-Bush effect that is statistically insignificant).
The popular media pronounced Gore the winner of the state of Florida before the polls had closed in 9 western counties in Florida, all of which are in the central time zone, and thus one hour behind the rest of the state. Bush supporters claim that this potentially hurt Bush’s vote count in Florida, which could turn out to be decisive (if Gore is eventually pronounced the winner in Florida) given the closeness of the popular vote. While yet to be published or confirmed by us, John Lott, an economist at the University of Chicago, is rumored to have evidence supporting Bush’s claim.
The potential effect is very similar to what happens in states on the pacific coast in presidential elections where the popular media has already pronounced an overall winner before the polls have closed. Voters who have yet to vote in these places now believe their vote will have no impact on the outcome of the election. In these cases two effects generally dominate: (1) voter turnout is less; and (2) voters are more likely to “waste” their vote on a third party candidate as a statement. It is unclear in advance (to us at least) which candidate would likely be hurt more in relative vote share from these effects. Both Bush and Gore voters would have less incentive to turn out to vote, and it is likely that Nader voters (who might have voted for Gore if the election was close enough) could offset any similar effect of Buchanan voters (who might have voted for Bush if the election was close enough). Bush supporters, however, are currently claiming that they are the side that has lost votes as a result of this central time zone phenomenon.
Here we explore this with three, fairly straightforward steps. First we see whether the vote margin between Gore and Bush is significantly different in these central time zone counties after controlling for other factors. Our results suggest that in these counties Bush received 4.26% larger margin of victory (difference in the popular vote percentages) than would have been expected. This result is statistically significant, but only at a 10% level of significance, which is not nearly as significant as one would hope to find to make any sort of convincing argument or to draw a firm conclusion. Thus, if any effect is present (which we don’t think there is based upon our other results), it seems that Gore (not Bush) was hurt in these counties in the central time zone. Perhaps a more clear way of stating this result is that we find no evidence that Bush’s total vote margin in Florida is smaller as a result of the media pronouncing Florida for Gore early in the evening.
To see whether further insights could be obtained by looking at the underlying factors, we then took a closer look at both voter turnout and third party vote share directly. After all, nowhere is it claimed that people switched their vote between Bush and Gore because of the media calling the election, but rather that it either lowered turnout or caused some voters to ‘throw their votes away’ by voting for a third party candidate. Our results do not find support for either of these effects. We cannot find any difference in voter turnout nor in the percentage of votes for third party candidates than should have been expected in these central time zone counties.
Our conclusion is that the popular media’s pronouncing Gore the winner of Florida before the polls closed in these 9 western Florida counties had no significant impact on the vote totals for either Bush or Gore, and thus would not have been a decisive factor in the election.
The detailed estimates from our models are given in the following section along with the variable descriptions.
In this model we employ standard OLS regression methodology to the following data at the county level for the state of Florida:
1) Dependent Variable: [Percent Vote for Gore – Percent Vote for Bush] * 100
2) Independent Variables:
Central Time Zone: Equal to one in counties in the central time zone
Black: Percent of county population black * 100
Hispanic: Percent of county population Hispanic * 100
Age65: Percent of county population age 65 or over * 100
College: Percent of county population with a college degree * 100
Per Capita Income: County per capita income (in thousands)
1996 Clinton/Dole spread: [%Vote for Clinton – %Vote for Dole] * 100
OLS Regression Results
|
|
|
|
|
|
Dependent Variable: Gore/Bush Spread [% difference in Gore/Bush
popular vote] |
||||
|
|
|
|
|
|
|
|
Coefficient |
Std. Error |
T-ratio |
|
|
|
|
|
|
|
|
Constant |
-50.325 |
|
|
|
|
Central Time Zone |
-4.260 |
2.489 |
-1.711 |
* |
|
% Black |
0.377 |
0.092 |
4.087 |
*** |
|
% Hispanic |
0.072 |
0.101 |
0.715 |
|
|
% Age 65 & over |
0.794 |
0.136 |
5.852 |
*** |
|
% College |
0.571 |
0.191 |
2.984 |
*** |
|
Per Capita Income (thousands) |
0.461 |
0.290 |
1.588 |
|
|
1996 Clinton/Dole spread |
0.900 |
0.066 |
13.684 |
*** |
|
|
|
|
|
|
|
R Squared |
0.898 |
|
|
|
|
# of Observations |
67 |
|
|
|
|
Degrees of Freedom |
59 |
|
|
|
|
|
|
|
|
|
|
Notes: *** indicates statistical significance at 1%, ** at 5%,
and * at 10% (two tailed) |
||||
In this model we employ standard OLS regression methodology to the following data at the county level:
1) Dependent Variable(s):
Model [1]: Total 2000 presidential votes as percent of registered voters
Model [2]: Total 2000 presidential votes as percent of 1996 presidential votes
2) Independent Variables:
Central Time Zone: Equal to one in counties in the central time zone
Black: Percent of county population black * 100
Hispanic: Percent of county population Hispanic * 100
Age65: Percent of county population age 65 or over * 100
College: Percent of county population with a college degree * 100
Per Capita Income: County per capita income (in thousands)
OLS Regression Results
|
|
|
|
|
|
Model [1] Dependent Variable: |
Total 2000
presidential votes as % of registered voters |
|||
|
|
|
|
|
|
|
|
Coefficient |
Std. Error |
T-ratio |
|
|
|
|
|
|
|
|
Constant |
25.251 |
|
|
|
|
Central Time Zone |
-1.862 |
6.473 |
-0.288 |
|
|
% Black |
0.028 |
0.234 |
0.120 |
|
|
% Hispanic |
-0.175 |
0.265 |
-0.661 |
|
|
% Age 65 & over |
0.823 |
0.362 |
2.275 |
*** |
|
% College |
0.694 |
0.490 |
1.417 |
|
|
Per Capita Income (thousands) |
0.763 |
0.722 |
1.056 |
|
|
|
|
|
|
|
|
R Squared |
0.257 |
|
|
|
|
# of Observations |
67 |
|
|
|
|
Degrees of Freedom |
60 |
|
|
|
|
|
|
|
|
|
|
Notes: *** indicates statistical significance at 1%, ** at 5%,
and * at 10% (two tailed) |
||||
OLS Regression Results
|
|
|
|
|
|
Model [2] Dependent Variable: |
Total 2000 presidential votes as % of 1996 presidential votes |
|||
|
|
|
|
|
|
|
|
Coefficient |
Std. Error |
T-ratio |
|
|
|
|
|
|
|
|
Constant |
71.060 |
0.000 |
|
|
|
Central Time Zone |
-8.642 |
12.068 |
-0.716 |
|
|
% Black |
-0.049 |
0.436 |
-0.113 |
|
|
% Hispanic |
-0.367 |
0.494 |
-0.742 |
|
|
% Age 65 & over |
1.007 |
0.674 |
1.494 |
|
|
% College |
0.811 |
0.913 |
0.888 |
|
|
Per Capita Income (thousands) |
0.747 |
1.347 |
0.555 |
|
|
|
|
|
|
|
|
R Squared |
13.837 |
|
|
|
|
# of Observations |
67 |
|
|
|
|
Degrees of Freedom |
60 |
|
|
|
|
|
|
|
|
|
|
Notes: *** indicates statistical significance at 1%, ** at 5%,
and * at 10% (two tailed) |
||||
In this model we employ standard OLS regression methodology to the following data at the county level:
1) Dependent Variable(s):
Percent vote for all third party candidates in 2000 * 100
2) Independent Variables:
Central Time Zone: Equal to one in counties in the central time zone
Black: Percent of county population black * 100
Hispanic: Percent of county population Hispanic * 100
Age65: Percent of county population age 65 or over * 100
College: Percent of county population with a college degree * 100
Per Capita Income: County per capita income (in thousands)
Third party percent vote in 1996 * 100
OLS Regression Results
|
|
|
|
|
|
Dependent Variable: |
Third party vote share * 100 |
|
||
|
|
|
|
|
|
|
|
Coefficient |
Std. Error |
T-ratio |
|
|
|
|
|
|
|
|
Constant |
1.281 |
|
|
|
|
Central Time Zone |
0.002 |
0.143 |
0.014 |
|
|
% Black |
-0.009 |
0.007 |
-1.236 |
|
|
% Hispanic |
-0.014 |
0.006 |
-2.174 |
** |
|
% Age 65 & over |
0.016 |
0.009 |
1.786 |
* |
|
% College |
0.064 |
0.011 |
6.003 |
*** |
|
Per Capita Income (thousands) |
-0.045 |
0.021 |
-2.166 |
** |
|
Third party vote share 1996 |
0.085 |
0.025 |
3.443 |
*** |
|
|
|
|
|
|
|
R Squared |
0.598 |
|
|
|
|
# of Observations |
67 |
|
|
|
|
Degrees of Freedom |
59 |
|
|
|
|
|
|
|
|
|
|
Notes: *** indicates statistical significance at 1%, ** at 5%,
and * at 10% (two tailed) |
||||