Some posters to this site have argued that leader or government approval ratings can be a better guide to general election results than the voting intention question, if not immediately before the vote, then early in the Parliament or in mid-term. I have been meaning for some time to put this to the test. I have used the IPSOS-MORI opinion poll and approval ratings data, which goes back to 1977, covering 11 general elections.
I have found:
- three years out, the government’s and the PM’s approval ratings are a slightly better predictor of the result of the next general election than the opinion polls, but the difference is fairly marginal and from a small sample size;
- two years away, no indicator is robust;
- in the last year of the Parliament, the government’s opinion poll lead is a better predictor of the result of the next general election than approval ratings;
- as you would expect, a low opposition leader’s gross approval rating just before the poll is associated with a government victory, while the PM’s gross approval rating does not seem to have a significant effect, except right before the election; and
- counter-intuitively, the government’s gross and net approval are better predictors three years out than they are just before the poll.
Those not interested in statistics can skip the remainder of this article.
Explanation of R2
The standard measure of statistical correlation is R2, generally expressed as a value between 0 and 1. An R2 of 1 indicates that the data perfectly fits the relationship expressed, in other words that all of the variation in one variable is associated with the variation in the other variable. An R2 of 0.5 indicates that half of the variation in one variable is associated with that in the other variable. And one of 0 indicates that they are entirely uncorrelated with each other.
As a methodology it has its problems, particularly with multivariable regressions, spurious correlations and small sample sizes. I have considered the R2 results with other measures (particular t-statistics and P-values for the regression coefficients) but these have generally agreed with the results below. I used MS-Excel’s standard regression package.
Generally, an R2 of 0.7 or higher (in bold in the results table below) indicates that the two variables are likely to be related, while one of less than 0.5 indicates that they may not be. Between those two numbers, the results are indeterminate.
The government’s share of the seats in Parliament at the following election since 1977 ranges from 25% won by John Major’s Conservatives in 1997 to 63.4% won by Tony Blair’s Labour in 2001.
I worked out the correlations between this variable and the seven following measures:
- Gross government approval
- Net government approval
- PM over Opposition Leader gross approval
- PM over Opposition Leader net approval
- PM gross approval
- Opposition leader gross approval
- Government opinion poll lead
I calculated these correlations for the following five lengths of time before the next election:
- Three years (this cuts the number of observations from 11 to eight, because the 2015 and 2019 Parliaments lasted less than three years, and the series only started in 1977, less than three years before 1979)
- Two years
- One year
- Six months
- The last survey.
This gives us 35 values of R2:
|Gross gvt approval||0.73||0.41||0.15||0.11||0.31|
|Net gvt approval||0.71||0.38||0.25||0.08||0.25|
|PM over Opp leader gross approval||0.57||0.40||0.42||0.63||0.79|
|PM leader over Opp leader net approval||0.65||0.45||0.32||0.51||0.73|
|PM gross app’l||0.71||0.26||0.03||0.05||0.25|
|Opp leader gross app’l(negative relationship)||0.08||0.54||0.60||0.69||0.78|
|Gvt opinion poll lead||0.57||0.53||0.78||0.71||0.90|
My thanks to the other PBers who have commented on the above.