¹ See, for example, Kellner, Peter and Worcester, Bob, ‘Electoral Perceptions of Media Stance’, in Worcester, R. and Harrop, M., eds, Political Communications (London: Allen and Unwin, 1982), pp. 57–67.Google Scholar

² Dunleavy, Patrick and Husbands, Chris, British Democracy at the Crossroads (London: George Allen and Unwin, 1985), p. 115.Google Scholar

³ Miller, William, Media and Voters (Oxford: Clarendon Press, 1991).Google Scholar

⁴ Miller, William, Clarke, Harold D., Harrop, Martin, LeDuc, Lawrence and Whiteley, Paul, How Voters Change (Oxford: Clarendon Press, 1990).Google Scholar

⁵ Miller, et al. , How Voters Change, p. 88.Google Scholar

⁶ Miller, et al. , How Voters Change.Google Scholar

⁷ Miller, et al. , How Voters Change.Google Scholar

⁸ Harrop, Martin, ‘Voters’, in Seaton, J. and Pimlott, B., eds, The Media in British Politics (Aldershot: Avebury, 1987).Google Scholar

⁹ Newton, Ken, ‘Do People Read Everything They Believe in the Papers? Newspapers and Voters in the 1983 and 1987 Elections’, Crewe, I. and Norris, P., eds, British Parties and Election Yearbook (London: Simon & Schuster, 1991).Google Scholar

¹⁰ Mosley, Paul, ‘“Popularity Functions” and the Role of the Media: A Plot Study of the Popular press’, British Journal of Political Science, 14 (1984), 117–29, p. 124CrossRefGoogle Scholar. It should be noted that our analysis of newspaper coverage is different from Mosley's. Mosley was essentially trying to compare officially recorded levels of, for example, unemployment with the levels reported by three Popular newspapers: the Mirror, Sun and Express.

¹¹ Mosley, , ‘“Popularity Functions” and the Role of the Media’, p. 128 (emphasis added).Google Scholar

¹³ The Independent was excluded from our sample because it was first published after the start of our sample period. The Financial Times was excluded on the grounds that its circulation was too limited.

¹⁴ Sanders, David, Ward, Hugh and Marsh, David, ‘Government Popularity and the Falklands Wan A Reassessment’, British Journal of Political Science, 17 (1987), 281–313.CrossRefGoogle Scholar

¹⁵ It was found that neither the inclusion of the editorial/non-editorial distinction nor the inclusion of information about the number of column inches (either in absolute terms or as a percentage of the total length of the paper) yielded results significantly different from those derived simply from examining the number of favourable versus unfavourable articles.

¹⁶ See Krippendorff, K. et al. , Content Analysis: An Introduction to Its Methodology (Beverly Hills. Calif.: Sage, 1980), pp. 130–54.Google Scholar

¹⁷ See Weber, R., Basic Content Analysis, 2nd edn (London: Sage: 1990), p. 17CrossRefGoogle Scholar

¹⁸ Krippendorff, et al. , Content Analysis, p. 132.Google Scholar

¹⁹ Full details of the reliability testing are available from David Marsh, Department of Government, University of Strathclyde.

²⁰ The particular smoothing algorithm employed throughout the present study was: smoothed Y ^t = (Y ^t + Y ^t–1)/2. Since only two time points are used, this method can be regarded as either ‘median’- or ‘mean-smoothing’.

²¹ The ‘non-Conservative’ papers covered in our analysis on this definition would be the Mirror and Guardian. We recognise that this distinction is not universally accepted (particularly with regard to The Times) and certainly would not claim to have justified it here. We are none the less convinced that it is sufficiently widely accepted both by other academic observers (particularly since Rupert Murdoch's takeover of Times Newspapers) and by the public at large to merit its investigation here.

²² In this context we were obviously dependent on officially-supplied employment statistics, the bases of which were repeatedly re-defined during the 1980s. Conveniently for our purposes, however, the main consequence of these redefinitions was to decrease the number of people classified as unemployed rather than to affect the variation in unemployment. As long as we used a consistent Series, therefore, our analysis was not likely to be contaminated by the effects of the repeated re*definitions. As a result, we simply used the most recently defined, fully back-calculated unemploy-mcnt series available at 1 January 1991.

It might appear that by using ‘back-calculated’ official statistics we are being ‘unfair’ to journalists who are obliged to rely on official statistics as they are published and before they are ‘revised’. (Mosley, ‘“Popularity Functions” and the Role of the Media’, p. 118Google Scholar). However, for the inflation, PSBR, exchange rate and interest series that we employ here, no revisions were made after initial Publication. Unemployment and balance of payments are the only series which are substantially revised. As we noted immediately above, revisions in unemployment figures do not affect the variation in this series – which is essentially what concerns us here. In addition, the official revisions to balance of payments figures are not problematic since we use exchange rate fluctuations in our empirical analysis.

²³ See Sanders, David and Ward, Hugh, ‘Art with Numbers: Timeseries Techniques for Repeated Cross-Section Data’, in Davies, Richard and Dale, Angela, eds, Analysing Social and Political Change: A Casebook of Methods (London: SageGoogle Scholar, forthcoming).

²⁴ See Sanders, and Ward, , ‘Art with Numbers’.Google Scholar

²⁵ Sanders, and Ward, , ‘Art with Numbers’.Google Scholar

²⁶ Given the high degree of co-linearity between the balance of payments and the external value of sterling, it was considered preferable to use only one of the two indicators as an exogenous variable. Since balance of payments figures are notoriously unreliable, we opted to use the exchange rate measures.

²⁷ PSBR figures are released on a quarterly basis only: lagging PSBR at t–3 and t–6 is accordingly equivalent to lagging the other objective economic indicators at t–1 and t–2.

²⁸ The specific measure used was the monthly Bank of England base rate.

²⁹ The PSBR and the exchange rate, both at time t, were also included in the list of potential independent variables at this stage. Neither proved significant in any of the ‘total economic coverage’ models.

³⁰ No election dummy for June 1987 is included. The election was held on 3 June and hence almost all of the press coverage during that month occurred after the election. Given that we are only concerned here with potential ‘run-up’ effects, therefore, we exclude June 1987 from analysis.

³¹ Only twenty-one observations out of a possible total of ninety-seven months (June 1979–June 1987) of data supplied by Gallup gave information on newspaper readership – all were between 1979 and 1983. We intend to undertake an individual-level analysis of these data at a later date.

³² Another possibility, of course, is that any correspondence between variations in Sun coverage and variations in popularity results from the fact that Sun coverage ‘reflects the prevailing mood of the electorate’. Even if coverage at time t–n correlates with popularity at time t, it can still be argued that coverage anticipates the national mood. In these circumstances it must be acknowledged that, whenever a correlation between popularity and the economic coverage of a particular newspaper is observed, we cannot formally eliminate the possibility of either ‘reflection’ or ‘anticipation’ from our analysis. However, although it is always possible to conclude that the coverage concerned has somehow reflected or anticipated the national mood, we would argue that in general Journalists are neither so prescient nor so empirically-driven in their reports and commentaries as to merit this interpretation. Certainly the evidence that we provide below on the rather weak connections between economic coverage and objective economic changes would suggest that the empirical underpinnings of press coverage, in the economic sphere at least, are not particularly strong. An additional reason for examining this kind of aggregated coverage is that the contents of Prominent national dailies are frequently familiar to more than just the individuals who purchase Particular newspapers. Not only are many newspapers read by more than one person, especially in the workplace, but most daily news and current affairs programmes on national radio and television feature a review of ‘today's papers’. We would argue that these factors together mean that many electors have a general sense of the overall balance of coverage that the press as a whole provides. It is this general sense of balance, in part, that our aggregated coverage variables Seek to explore.

³³ Sanders, and Ward, , ‘Art with Numbers’Google Scholar; Marsh, , Ward, and Sanders, , ‘Modelling Government Popularity in Britain’Google Scholar. The ‘Misery Index’ that we reported as being significant in the latter article was not included in the core popularity equation on the grounds that it appears not to have been so important empirically in the period since 1987.

³⁴ Though we do not report the results here, none of the category-by-category coverage variables yielded a significant coefficient when added to the ‘core’ model: this is why Table 10 only shows the effects of the ‘overall economic coverage’ variables identical to those specified in Table 7. It is also worth noting that the addition of the various press coverage variables to the core model Calces very little difference to the coefficients of that model, suggesting that the core model is highly stable.