Laying the Foundations for Electoral Success: Conservative Pre-Campaign Canvassing before the 2010 UK General Election

Abstract

The Conservatives needed a net gain of 116 seats at the 2010 general election if they were to win an overall majority and form the next government. From 2007 onwards, a unit in the central party organization, led by Conservative Deputy Chairman Lord Ashcroft, worked with local parties to promote their candidates' cause in a substantial number of marginal constituencies. The efforts of Lord Ashcroft's team involved the expenditure of several million pounds during the pre-election campaign period in an effort to win over voters in key battleground seats. But was it effective? Using a path modelling approach, we provide substantial evidence that it was. Not only did it have both a significant direct effect and an indirect effect, through “short campaign” effort, on Conservative Party support in 2010, but the Conservatives were significantly more likely to win key battleground seats against Labour where larger grant allocations were made. Frequency also mattered. The Conservatives were far more likely to win Labour-held seats if constituencies received money from Lord Ashcroft's team on two or more occasions.

Notes

Lord Ashcroft's target seat programme took place over three years (2007, 2008 and 2009). We categorize this as the pre-election campaign period. The “long campaign” refers to activism in the run up to the election (roughly four months prior) whereas the “short campaign” reflects campaigning during the 3–4 week election period.

One of us (Ron Johnston) had a long telephone discussion with one of the leading members of the team, Stephen Gilbert, in summer 2010.

Full details of the accounts can be found on the Electoral Commission website at <www.electoralcommission.org.uk/party-finance/party-finance-analysis/party-funding>.

That this is such a common practice suggests that it was party policy not to have local parties draw attention to their receipt of these grants, but this has not been confirmed.

Nearly all the grants stated went to local Conservative parties in England.

Because there were new constituencies in England and Wales, all of the analyses here use the estimated outcome of the 2005 election if it had been held in the new constituencies (Rallings & Thrasher, 2007).

For additional clarity, it is important to stress that while Lord Ashcroft established his target seat strategy within the central Conservative headquarters and undoubtedly contributed money to the party (through his company Bearwood Services Ltd) it was Conservative money that was spent. To reflect this, the use of the term “Ashcroft spending” is placed in inverted commas.

The Ashcroft variable uses raw spending data, however, we log transform this due to evidence of skewness and some minor kurtosis.

For the “short campaign”, the maxima allowed were: in a county constituency, £7,150 plus 7p for every entry in the register of electors in the constituency on the due date; and in a borough constituency, £7,150 plus 5p for every entry in the register of electors in the constituency on the due date.

Expenditure during the “long campaign”: in a county constituency, £25,000 plus 7p for every entry in the register of electors; in a borough constituency, £25,000 plus 5p for every entry in the register of electors. In 2010, the median constituency electorate was 70,189: its maximum allowed expenditure for the short campaign was £12,063.23 for a county constituency, and £10,659 for a borough, and for the long campaign the limits were £29,913.23 and £28,509.45 respectively (see Johnston et al., 2011).

There is a case to include casual arrows from both prior party vote and party incumbency to reflect the possibility that they may have informed “Ashcroft spending”. However, we are not seeking to measure what informed “Ashcroft spending”. Our primary focus is the consequences of “Ashcroft spending” on party support (both directly, and indirectly through campaign spending). The inclusion of such pathways is not likely to affect these relationships hence our decision to exclude them from the model.

During the modelling process, we used the full sample of 632 constituencies. Where there was missing data, we used the estimation-mobilization (EM) algorithm in Mplus to compute missing data estimates using full information maximum likelihood (FIML) (Muthén et al., 1987). This preferable to listwise deletion, pairwise deletion, mean substitution and other missing data solutions because it has produces unbiased parameter estimates and standard errors under Missing at Random (MAR) and Missing Completely at Random (MCAR). To obtain reliable model fits, the FIML method requires incomplete data to be at least MAR (Little & Rubin, 1987). To test this we modelled missingness using the covariates included in the model. The results showed that any data was randomly missing conditional on other variables in the model.

The standard goodness of fit measures indicated a good model – both the Comparative Fit Index (CFI) and the Tucker Lewis Index (TLI) was 0.95 or above, and the Root Mean Square Error of Approximation (RMSEA) was less than 0.05 (Hu & Bentler, 1999).

Some of the grants were only small because the local party was relatively well-resourced and didn't ask for as much in their business plans.

It is important to note that we ran the same models using the Ashcroft variable as a binary measure and got the same results. These findings are available on request.

We also re-ran the models to take account of the MPs expenses scandal on party support. First, we included dummy variables to represent where parties were “tainted by the scandal”. Second, we then split this to take account of whether a party incumbent stood down or decided to stand for re-election. In all cases, these MP expenses variables were not significant at the 95% confidence level (the general expenses measure for Labour was negative and significant at the 90% level). Crucially, they did not affect the direct effect, and indirect effect through campaign spending, of “Ashcroft spending” on Conservative Party support. Given these findings, we decided to omit these variables from the model.

The finding that Conservative spending during the “official campaign” period boosted Liberal Democrat support seems at first glance to be counterintuitive. However, we would argue that stronger Conservative campaigning in seats where the Liberal Democrats were their main rivals could have conceivably driven more Labour supporters into voting tactically for the Liberal Democrats. Hence the finding that Liberal Democrat support increased the harder the Conservatives campaigned during the “official campaign” period.

It was not possible to run a mediator model to differentiate between the direct and indirect effect of Ashcroft spending on whether the Conservatives won the Labour-held seats or not. This is because such models are not run on binary outcomes. Instead we run a logistic regression path model in Mplus using the Monte-Carlo algorithm and maximum likelihood estimation with robust standard errors.

Marginality is operationalized as a binary variable (1 = <20% behind the winner in 2005; 0 = >20 behind the winner in 2005). We tested the model with other measures of marginality e.g. <15%; <10% or including categories of margin (ultra-safe, safe, fairly safe, fairly marginal and ultra-marginal) and found very similar results. We, therefore, used a binary variable for ease of interpretation.

Twenty constituencies made up category one, while 28 seats received assistance from Lord Ashcroft on one occasion in 2009. In total, 34 constituencies obtained Lord Ashcroft money for two years or more over the three-year period.