//
you're reading...

Datablog

“It’s (not) the economy, stupid”: How economic indicators swing pundits not American presidential elections

Today Obama will have the inauguration ceremony for his second term; overcoming what seemed to be the impossible ailment of a stagnant economy. In his election year unemployment stood at 8.1%, unseen in America since the end of the great depression; this was coupled with GDP rising at a rate that only Europe would envy – 2.1%. On the surface Obama defied the odds to win the election. Pundits declared the economy as Mitt Romney’s ‘Open Goal’, that Barack Obama’s poll ratings were directly aligned with the health of the economy, even Bill Clinton noted that voters were impatient for recovery; and its easy to understand why. More worrying for Obama: in most polling, Americans listed the economy as the most important issue as they went to the ballots.

data consistently suggests that economic growth and particularly unemployment does not predict or decide presidential elections

Despite his economic record Obama still won, contrary to much of the election commentary from pundits. This was not necessarily due to Mitt Romney’s gaffe-prone candidacy, or because Obama won on other issues. It is because the economic indicators had little to no sway over the election. Data consistently suggests that economic growth and particularly unemployment does not predict or decide presidential elections. If nothing else this post should show that Barack Obama’s election win as an incumbent with poor/weak economic indicators is far from unusual.

The graphs below show’s all presidential candidates who have run as the current President or Vice-President, so only candidates who can reasonably be considered to be to some degree responsible for the state of the economy at the time of election. The first shows both the GDP growth and unemployment rate in their election year by president in order of their popular vote, the second shows the relationship between GDP growth and the candidates popular vote whilst the third does the same for unemployment and popular vote.

Hovering over data points shows further information on the candidate including their election year.

In the first graph there is no obvious, consistent relationship between unemployment, GDP growth and the candidates popular vote; in some areas it is highly inconsistent. The relationship between the individual variables GDP and Unemployment (the second and third graphs) do show the relationship that would be expected: as GDP increases and unemployment falls the incumbent candidates popular vote increases. Yet the manner in which the data points are dispersed around the regression line shows just how weak the relationship is. When treated individually unemployment has a 51% probability of being completely insignificant to the overall popular vote; while GDP fairs considerably better, its reliability is very weak.

treated individually unemployment has a 51% probability of being completely insignificant to the overall popular vote

Looking at some of the data points the weakness associated with unemployment is not surprising. Obama has had the highest unemployment of any incumbent who has run for president since 1944 yet still had a popular vote above that of George Bush’s election in 2004. Reagan in 1984 won in a landslide with 7.4% unemployment; while Hubert Humphrey in 1968, Al Gore in 2000 and Bush Snr. in 1992 lost their election with unemployment levels below the historical average. Additionally, I have only gone back as far as the data provided by the US Bureau of Labor Statistics and Bureau of Economic Analysis allows. FDR’s 1936 re-election came when unemployment was above 16%, he then followed this by getting re-elected once more with similarly unfavourable indicators.

GDP growth is a much more reliable figure, with only a 7% probability of being insignificant. All this means is that economic growth is likely to affect an election, but the extent to which it does is far from certain and highly variable. So the relationship found in the data, that every percentage increase in GDP will increase the incumbents vote by 1.5%, will rarely hold true.

Nate Silver, the New York Times blogger who successfully predicted the 2012 election in each state, also noted that economic variables have no particular relationship with US Presidential elections– and used it as his justification for not changing his prediction as new economic data was announced. Gerald Kramer of Yale University did one of the major studies into the impact of economic data upon US elections, also found that Presidential elections are “substantially less responsive to economic conditions” than congressional elections which can fluctuate with economic variables. If nothing else that should provide some form of hope that the legislative body chosen by the electorate that decides the budget usually has some form of relationship with economic data.

The factors that contribute to an American Presidential election, as well as the factors that produce economic indicators are numerous, hard if not impossible to measure, and require rigorous analysis. In an election where fact-checking was a key part of the narrative, it seems incongruous that employment figures and GDP played the significant role that it did. In that sense Barack Obama did not buck the trend winning an election in a fraught economy, but simply continued the trend of punditry not representing the actions of voters.


Data and Methodology

Presidential Winners Incumbents from previous administration (inc. VP’s)
Winner GDP Change Unemployment Their Popular Vote Year Candidate GDP Unemployment Popular Vote Year
Franklin D. Roosevelt 8.08% 1.20% 53.39% 1944 Franklin D. Roosevelt 8.08% 1.20% 53.39% 1944 Won
Harry S. Truman 4.40% 3.80% 49.55% 1948 Harry S. Truman 4.40% 3.80% 49.55% 1948 Won
Dwight D. Eisenhower 3.83% 3.00% 55.18% 1952 Dwight D. Eisenhower 1.98% 4.10% 57.37% 1956 Won
Dwight D. Eisenhower 1.98% 4.10% 57.37% 1956 Richard Nixon 2.48% 5.50% 49.50% 1960 Lost
John F. Kennedy 2.48% 5.50% 49.72% 1960 Lyndon B. Johnson 5.79% 5.20% 61.05% 1964 Won
Lyndon B. Johnson 5.79% 5.20% 61.05% 1964 Hubert Humphrey 4.84% 3.60% 42.70% 1968 Lost
Richard M. Nixon 4.84% 3.60% 43.42% 1968 Richard M. Nixon 5.31% 5.60% 60.67% 1972 Won
Richard M. Nixon 5.31% 5.60% 60.67% 1972 Gerald Ford 5.36% 7.70% 48.00% 1976 Lost
Jimmy Carter 5.36% 7.70% 50.08% 1976 Jimmy Carter -0.28% 7.10% 41.00% 1980 Lost
Ronald Reagan -0.28% 7.10% 50.75% 1980 Ronald Reagan 7.19% 7.50% 58.77% 1984 Won
Ronald Reagan 7.19% 7.50% 58.77% 1984 George Bush 4.11% 5.50% 53.37% 1988 Won
George Bush 4.11% 5.50% 53.37% 1988 George Bush 3.39% 7.50% 37.50% 1992 Lost
Bill Clinton 3.39% 7.50% 43.01% 1992 Bill Clinton 3.74% 5.40% 49.23% 1996 Won
Bill Clinton 3.74% 5.40% 49.23% 1996 Al Gore 4.14% 4.00% 48.40% 2000 Lost
George W. Bush 4.14% 4.00% 47.87% 2000 George W. Bush 3.47% 5.50% 50.73% 2004 Won
George W. Bush 3.47% 5.50% 50.73% 2004 Barack Obama 2.10% 8.10% 51.10% 2012 Won
Barack Obama -0.34% 5.80% 52.87% 2008
Barack Obama 2.10% 8.10% 51.10% 2012

 

All of the regressions used Ordinary Least Squares, this finds the line of best fit that has the smallest distance possible from all of the data points. This is found by minimizing the square of the total distance between each of the points and the line of best fit. The relevance of each variable such as GDP was measured either using the F-statistic or t-statistic, and the amount the found relationship varies was measured using the co-efficient of determination (also known as R2 statistic).

I should note that this is a relatively simplistic regression analysis, the Kramer paper has a much more thorough and accurate methodology.

Regression between Popular Vote and GDP growth

Relationship:
PopularVote% = 48.1% + (1.498 × GDPgrowth%)

t-statistic for GDP coefficient:
1.922, which gives a 7.5% probability of insignificance

R2 statistic:
0.208

Note: all R2 statistics vary between 0 and 1, where 1 represents a relationship that fits all datapoints perfectly.

 

Regression between Popular Vote and Unemployment

Relationship:
PopularVote% = 54.25% + (-0.639 × Unemployment%)

t-statistic for GDP coefficient:
-0.668, which gives a 51.4% probability of insignificance

R2 statistic:
0.031

 

Bivariate Regression between Popular Vote and GDP + Unemployment

Relationship:
PopularVote% = 44.7% + (-0.027 × Unemployment%) + (1.489 × GDPgrowth%)

F-statistic for GDP and unemployment coefficients:
1.716, which gives a 21.8% probability of insignificance

R2 statistic:
0.208

 

Discussion

No comments yet.

Post a Comment