Political advertising – a brief survey
The empirical political advertising literature is limited. Most authors refer to Rothschild’s seminal (1978) article, which is both a review and a report on his tests of an involvement model. The level of involvement of voters in a particular election could depend on a wide range of factors, but Rothschild offers three general ones: 1. the level of the election (national/local etc.); 2. the closeness of the race; and 3. the volatility of the issues or candidates.
There is some more recent work examining voter recall of political messages. Faber and Storey (1984), note voter responses to different messages and presentations (see, for example, Thorson et al., 1991), in particular negative ones (Faber et al., 1993; Garramone, 1984; Tinkham and Weaver-Lariscy, 1994). These studies used subjects in experimental groups rather than actual electoral or opinion poll data. More recently Ansolabehere and Iyengar (1995) have substantially added to the literature on negative advertising through the use of a very rounded and full quantitative and qualitative study which shows that a highly focused negative campaign can increase the turnout of those potential supporters predisposed to abstain by something in the order of approximately 29 per cent, thus maximising your supporters’ vote.
Weaver-Lariscy and Tinkham (1987) investigated how Congressional election candidates evaluated different media and proposed six levels of response to political advertising. Apart from Caywood and Preston (1989) and Laczniak and Caywood’s (1987) work with others cited above, a number of authors have considered the ethical issues involved (Kaid, 1991; Tinkham and Weaver-Lariscy, 1994). Rothschild (1978) raises the public policy issues involved in political advertising and this theme is taken up by other authors (e.g. Palda, 1994; Richards and Caywood, 1991). Magleby (1994) in one of the first studies in the area outlines his preliminary findings on the relative effectiveness of communication tactics in political campaigns in referendums (California Initiative Process), he suggests targeted press campaigns are deemed important in influencing differential turnout.
There is, however, only a limited amount of work, which has been used to estimate directly electoral response to political advertising. Palda (1975) considered the impact of advertising and other factors on voting behaviour in two Canadian provinces (Rothschild considered Palda’s studies to be the only ones based on actual voting behaviour). Palda found that advertising level affected voter turn-out. Rothschild’s experiments suggested that the effect of advertising depended on whether involvement was high or low. In low involvement races political advertising had a strong positive effect on voting intention, but not in high involvement ones. Pommerehne and Lafay (1983) presented an overview of econometric work analysing expenditure. Nearer home, Atkin and Heald (1976) offered an analysis of the effects of political advertising (Gordon Heald was also called as an expert witness in the Irish High Court case).
Interestingly Demick et al. (1996) have reviewed Public Service Advertising in Northern Ireland and discuss in depth the impact it has on social issues. They discuss the above in the context of its own conceptual framework which is built upon a longitudinal advertising campaign run by the Northern Ireland Office over a period of eight years. The overall purpose of the campaign was to publicise the confidential telephone system and to stimulate change in society with regard to terrorism and the need for ordinary people to play their part. The confidential telephone system, which was established to counteract intimidation, enables the public to contact the police anonymously through an answering machine connected to a free phone (0800) number. The importance of this study is that when the public service advertising has been shown on TV, responses have been logged and the different appeal rates of each advertising campaign have been tracked.
It should also be noted that press advertising can play a key role in political agenda setting. In the 1979, 1983 and 1987 elections, press advertising played a significantly influential role, both in agenda setting and supporting the main campaign theme.
Anyone familiar with the literature on the advertising sales relationship for convenience products will be familiar with the problems of estimation encountered (see, for example, Blattberg and Jeuland, 1981). The political advertising measurement problem is considerably greater. First, for any particular election or referendum there is really only one electoral data point, polling day. Of course we have opinion polls, though these are based on comparatively small samples, which makes cross-sectional as well as longitudinal work difficult as it is rarely possible to match variations in sub-samples with different advertising levels and messages.
Finally, because of coverage in other media, there is potentially much more noise in the data. From the literature, we hypothesise that the Irish Divorce Referendum should be considered a high involvement election and that as a consequence, there should be a significant impact of advertising on turn-out, but only a limited effect on voter preferences.

The impact of political advertising

For the purposes of the analysis in this article, we are considering advertising immediately preceding an election – in this case, a referendum. We therefore consider two separate effects – one the effect of advertising on the propensity to vote, and the second on which way the individual chooses to vote. It should be noted that propensity to vote effects may not be symmetric – a particular campaign may have a greater effect on the opposition’s propensity to vote than on that of one’s own side’s supporters.
For the purposes of referendum with one question with a simple binary response, we define, at any point in time t, three classes of voters: 1. those who would vote yes, if they voted – Y(t); 2. those who would vote no, if they voted – N(t); and 3. those who do not know (or care?) – DN(t).
All the variables are defined as proportions of the total numbers eligible to vote, and can be extended by a further subscript (i) to define the related to electoral area i, at whatever level of aggregation. For the first two groups, we also define P(t)|Y(t) and P(t)|N(t), which are the conditional propensities to vote of the two decided groups. We have not defined P(t)|DN(t) in this situation as the proportion of blank or spoiled ballot papers were insignificant. If we define E(i) as the total electorate in area i, then the figure for those voting yes at an actual referendum at time t, is given by E(i).P(t)|Y(i,t).Y(i,t), and similarly for the noes, E(i).P(t)|N(i,t).N(i,t). It should be noted that neither the underlying proportion or the conditional propensities to vote are usually directly observable. Whilst we know E(i) (though one recognises that electoral registers in areas with mobile populations “age” rapidly), the actual observations we have are: YES(i) as the number voting yes in electoral area i, and similarly NO(i) for the noes.
By contrast, opinion polls at some time (t-x) yield estimates of Y(i,t-x), N(i,t-x), and DN(i,t-x), which may or not be corrected by estimates of the relative voting propensities of the different groups. Fortunately for this analysis, the Irish have a reputation of being relatively truthful in opinion polls. (This might be contrasted with the continuing post-mortem over the opinion poll “failure” at the 1992 UK General Election, JRSS, 1996, and Worcester, 1996.)
We propose, therefore, that changes in the propensity to vote in the immediate run-up to a referendum (t-1) (P(t)-P(t-1) and the relevant conditional probabilities) are affected by advertising in the time period. The impact on preferences are more complex as may be seen from Equations 1.1, 1.2 and 1.3:(see equation 1) (see equation 2) (see equation 3) (see equation 4) (see equation 5) (see equation 6)Advertising can have an effect on each of the switching elements, though these are not directly observable, though are sometimes tracked or estimated through small longitudinal panels. This relatively simple model does not take into account relative strength of views within each group. In addition because we are only considering one advertisement in multiple newspapers, the effectiveness of the actual copy cannot be distinguished from the spend, regional and coverage effects.