Supplement to “ What Drives Media Slant ?

TABLE B.I PRESENTS THE QUANTITIES shown in Table IV of the paper for a series of alternative model specifications. For comparison, Table B.I first reports the results in Table IV as the baseline. The first specification in Table B.I allows each zip code to contribute differently to a newspaper’s profits, as discussed in Section 4.3. We use crossnewspaper variation in ad rates to estimate the approximate advertising value of different demographics. We then apply the coefficients from this regression to each zip code’s demographic characteristics to produce an estimate of the advertising value of readers in each zip code. Finally, we combine these estimates with data on cover prices and estimates of marginal costs (following Gentzkow (2007)) to generate a per-reader markup for each zip code. We then use the sum of the markup-weighted number of readers in each zip code as the maximand when computing idealn for each newspaper. The second specification of Table B.I excludes newspapers headquartered in multi-paper cities from our supply sample, as discussed in Section 4.3. The third specification of Table B.I uses an alternative slant measure with a different selection of partisan phrases, as discussed in Section 3.1. In particular, we tighten the cutoffs on the number of hits a phrase must have in newspaper headlines from 2000 to 2005 by setting them equal to the 5th and 95th percentiles in our sample. We then select from the remaining phrases the top 1000 by χpl. The fourth specification of Table B.I uses an alternative slant measure computed by taking a congressperson’s adjusted Americans for Democratic Action (ADA) score (Groseclose, Levitt, and Snyder (1999)) as a measure of ideology, as discussed in Section 2.1. The last specification of Table B.I uses an alternative slant measure computed by taking a congressperson’s common space NOMINATE score (Poole and Rosenthal (1997, 2007), Poole (1998)) as a measure of ideology, as discussed in Section 2.1. Note that because the last two specifications use different measures of congressional ideology, the units of slant measure are different. Up to the change in units, our core results hold in all specifications: actual and profit-maximizing slant levels are similar on average, deviations from profit maximization result in an economically significant loss in profit, variation in consumer ideology explains a significant portion of the variation in slant, and variation in owner ideology explains a small portion of the variation in slant.