The leadership of the major parties always seems to be a constant source of speculation in the political universe. The Nielsen poll released on Monday added to this speculation by canvassing alternative leadership options for both major parties. The alternatives were not novel being the former leaders of each party, Kevin Rudd for Labor, Malcolm Turnbull for the Coalition.
A more prominent focus was given to a Rudd return to the Labor leadership and The Lodge. A voting intention question was asked on a hypothetical Rudd leadership of the Labor party. The results were not unexpected given that the same question has been asked previously by the same firm.
In Monday’s poll the current Labor primary vote rose from 34% under the usual question (and the current leadership) to 45% under the hypothetical Rudd leadership. On a TPP basis the ALP vote rose from the current 47% to 53%, a 6% rise. This compares to a rise of 10% (Labor TPP under Gillard 42%, Rudd 52%) in September 2011 and a more sanguine 3% gap measured by Galaxy prior to the leadership challenge in February (Labor TPP under Gillard 46%, Rudd 49%)
There was no similar voting intention question for a hypothetical Turnbull Coalition leadership in Monday’s Nielsen poll.
However, are such questions of voting intention under alternative leaders reliable? Voting intention questions with hypothetical leaders have a cloud over them as they, by necessity, require prompting by the pollster of the novel leadership option in the way the question is asked. Such prompting is not included in normal voting intention questions and in any event the mere novelty of the new leader may prompt some swinging voters to change for the sake of it. Furthermore we know one thing for certain about hypothetical leader, in that they do not have the baggage of current incumbency.
All these issues cannot be filtered out of polling questions featuring a hypothetical leader, they are part and parcel of the exercise. In other words caveat emptor.
So is their another way we can measure the effect on party support of changing leaders?
Measuring polling effects of recent leadership changes.
We can try the obvious exercise of comparing opinion poll performances immediately prior to and after mid-term changes in major party leadership. We shall avoid leadership changes that take place immediately after elections as they take place in the post-election victory glow and we are focusing on the effect of mid-term changes any way. We shall therefore consider a sample of 10 mid-term leadership changes of the major parties since 1989.
For our exercise today we shall Newspoll, due to availability of its records over the past 30 years and its constant polling over the same time frame. It is the only pollster with available records that has polled frequently soon after leadership changes, many of which occur in late November of December. Newspoll with its fortnightly polling is often the only poll to report after that time. In any event a caveat for this exercise is that we are using a single pollster.
To avoid any coincidental outlier we are going to use an average of the last 3 Newspolls prior to the Leadership change and the next 3 Newspolls after the leadership change as the basis of our before and after comparison. We shall measure the change in voting intention on a TPP basis. The TPP results will be calculated on the basis of preference flows at the previous election applied to the Newspoll primary vote figures.
We are also going to look at the change in the leadership favourability ratings. Normally these questions are a bit of a beauty contest as they are not a direct indicator of voting intentions and it is the votes which decide elections. But we will see if there is any relationship between the change in leadership favourability ratings and the TPP in the vicinity of the leadership change. As a side note we have used the term favourability in this post. This is interchangeable with the term satisfaction, the latter term is used by Newspoll.
We will assess the change in favourability ratings by calculating a modified leadership favourability score and comparing the outgoing leader’s score with the incoming leader’s to create a swing.
A note on the modified leadership favourability score
The formula of the modified leadership favourability score is as follows: (fav-unfav) x (100-uncommitted)
In short is if the leadership net favourability weighted by the share of respondents who are uncommitted. Once again we are taking the average of over a 3 poll period immediately prior to and after the changes.
Data and scatterplots
The first take home message is that the TPP moves between -0.66% to 4.33%. A sharp move of 6% on a TPP basis to Labor as indicated by the Nielsen poll for a Rudd Labor leadership would be historic compared to recent leadership changes.
Does the relative popularity change matter of the leaders matter? Let’s do a scatter plot.
It seems to. The larger the swing in the modified leadership favourability, the larger the short term TPP swing.
If we did a quick back of the hand calculation in accordance with the above relationship, a new leader who had a favourability rating of 50-40-10* (Fav-Unfav-Unc) would score a 44 point swing against the current leaders’ modified leadership favourability swing which has been hovering around 30-60-10. This would translate into a 2% jump in TPP based on the above relationship. Add a variability band around this of +/- 2% and we get a range of 0-4%.
*(which is our guesstimate of what the favourability ratings for Rudd and Turnbull would be after a reinstalment as leader)
But is this the whole story? Let’s plot another scatter plot comparing the change in the TPP vote after the leadership change with the TPP level prior to the change. The plot is below.
We get a graph with the same characteristics as the leadership favourability plot. The higher the pre-existing TPP, the higher the change in the TPP after the change in leadership.
A nerdy concern
So which variable (popularity or pre-existing TPP) drives the TPP change more? Well this is tricky because there is a mild positive relationship between the modified leadership favourability swing and the pre-existing TPP. Therefore there is a degree of multi-collinearity if we were to do a multiple variable regression with these two metrics as explanatory variables. In any event a multiple regression of the two metrics return R squared figures which are not significantly greater than the individual linear regression above. In short the two metrics tell us not much more together than they do separately.
Chicken or the egg ?
Simply put we are at a chicken or the egg scenario. Are voters pre-prepared to be favourable to a new leader given the current underlying political situation? Or does popularity of a new leader matter? We can’t tell one way or another so we will have to say it is both factors in tandem for now.
Based on the above analysis of historical changes in polling figures after previous leadership change we can say the following with the appropriate caveats:
- We are looking at a 0 – 4% bounce on a TPP basis If a new leader is chose for either of the major parties. This bounce is for the first 6-8 weeks after the change to the leadership. After that point other factors return to affect the dynamic of public support.
- This not to say the new leader will not have better polls immediately after their leadership victories. The above assessment is an average bounce over the 6-8 week short term period.
- Short term bounce isn’t everything. The best three leadership changes in term of a poll boost were; Downer 1994, Howard 1995 and Rudd 2006. You can pick the odd one out.
- As the Coalition are currently doing better in the polls with a 53.6% TPP according to the current poll average, a change of Coalition leader will obtain a higher bounce than a change of Labor leader.
The necessary caveats are as follows:
- We only have 10 observations to draw from.
- Records are made to be broken.
- We could only use one polling company; Newspoll. However we effectively took snapshots of polls over a 21 year period, so any house effect should be diluted.