There is mounting evidence that media coverage on election campaigns matters. Previous research has convincingly shown that opinions about parties, candidates and issues, the determinants of voting decisions, as well as the actual voting decisions are directly or indirectly affected by media coverage. However, most studies drawing on content analysis and survey data for the purpose of investigating effects of campaign coverage on voters' perceptions, attitudes or voting decisions, are either conducted on the aggregate level or resort to classical regression analysis on the individual level. So far, research that includes both perspectives of analysis is very seldom. Even less common is research, that systematically compares the results originating from one or the other strategy of analysis and discusses the theoretical and methodological consequences of these different perspectives. Moreover, although multi-level analysis has become more and more popular within the social sciences over the past decades, yet it has not been exploited for combining media and survey data in research on campaign or media effects in general.

By the means of a study about campaign coverage on candidates and its effects on voting decisions, we will investigate media effects on voters on all of the above mentioned levels of analysis. Drawing on content analysis as well as rolling cross-section survey data from a research project on campaign dynamics in the run-up of the 2005 German national election as well as comparable data conducted by the German Longitudinal Election Study (GLES) on the election of 2009, we examine whether the portrayal of the chancellor candidates in television news coverage influenced voters' perceptions and attitudes towards these candidates as well as their voting decisions during the 2005 and 2009 campaigns. For this purpose we use time series analysis on the aggregate level in the first step, followed by corresponding regression analysis on the individual-level. Finally, we explore the properties of multi-level models and discuss the assets and drawbacks of the different levels of analysis. Our results indicate that a multi-level approach is the most adequate way of linking media and survey data and analyzing effects of media coverage on voters.