C. Robustness In this section, we subject our tests for advertising bias to four robustness checks. Our primary concern is that own-publication advertising may proxy for unobserved fund quality or demand for information from the publication,s readers. First, in unreported tests, we find that controlling for future fund returns(measured from month t through month t+ 11)has relatively little impact on the magnitude or statistical significance of the coefficients on own-publication advertising for any of the publications in our sample. These findings are consistent with our finding in Section IV that recommendations are not informative about future returns and suggest that own-publication advertising does not proxy for unobserved fund quality. I Second, we re-consider the possibility that past advertising in a specific publication reflects (or in- creases)the demand of that publications readers for articles on advertisers'funds. 3 To address this concern we re-estimate the specifications in Table Ill, instrumenting past own-publication advertising in personal fi- nance publication k with past advertising in the two other personal finance publications. Advertising in these other publications is a valid instrument if it is related to a families' general propensity to advertise in personal finance publications but not to any unobservable preferences of publication-k readers for its funds. We follow Angrist and Krueger's[2001]advice and switch to a linear probability specification to avoid the difficulties associated with estimating limited-dependent-variable Iv models. In Panel A of Table Iv,we eport the coefficients on own-publication advertising for both Ols and iv. The Ols results are qualitatively similar to those in Table Ill, with positive and significant coefficients for the positive mentions and negative and insignificant coefficients for the negative mentions. The IV results are similar to those obtained using OLS, which suggests that own-publication advertising does not proxy for publication-specific interest in advertisers' fund Third, to control for observable variables in a less parametric manner, we re-estimate equation(1) 12We thank eferee for suggesting this robustness check GEorge and Waldfogel[2003]present evidence that r content responds to the demographic ers within their market. Similarly, Hamilton and Zeckhauser [2004, p. 5] find that media coverage of CEOs is increasing in the number of common shares outstanding and conclude that this reflects "the desire of reporters to write about firms with a wide audien of investors. These papers suggest that tests for advertising bias need to consider the possibility that both advertising and content respond to underlying subscriber demands .4This identifying assumption is subject to a criticism ar prices from other markets [Bresnahan 1997. If certain families persistently app cables(including the general level of print advertising), then othe is not a validC. Robustness In this section, we subject our tests for advertising bias to four robustness checks. Our primary concern is that own-publication advertising may proxy for unobserved fund quality or demand for information from the publication’s readers. First, in unreported tests, we find that controlling for future fund returns (measured from month t through month t + 11) has relatively little impact on the magnitude or statistical significance of the coefficients on own-publication advertising for any of the publications in our sample. These findings are consistent with our finding in Section IV that recommendations are not informative about future returns, and suggest that own-publication advertising does not proxy for unobserved fund quality.12 Second, we re-consider the possibility that past advertising in a specific publication reflects (or in￾creases) the demand of that publications’ readers for articles on advertisers’ funds.13 To address this concern, we re-estimate the specifications in Table III, instrumenting past own-publication advertising in personal fi- nance publication k with past advertising in the two other personal finance publications. Advertising in these other publications is a valid instrument if it is related to a families’ general propensity to advertise in personal finance publications but not to any unobservable preferences of publication-k readers for its funds.14 We follow Angrist and Krueger’s [2001] advice and switch to a linear probability specification to avoid the difficulties associated with estimating limited-dependent-variable IV models. In Panel A of Table IV, we report the coefficients on own-publication advertising for both OLS and IV. The OLS results are qualitatively similar to those in Table III, with positive and significant coefficients for the positive mentions and negative and insignificant coefficients for the negative mentions. The IV results are similar to those obtained using OLS, which suggests that own-publication advertising does not proxy for publication-specific interest in advertisers’ funds. Third, to control for observable variables in a less parametric manner, we re-estimate equation (1) 12We thank an anonymous referee for suggesting this robustness check. 13George and Waldfogel [2003] present evidence that newspaper content responds to the demographic mix of consumers within their market. Similarly, Hamilton and Zeckhauser [2004, p. 5] find that media coverage of CEOs is increasing in the number of common shares outstanding and conclude that this reflects “the desire of reporters to write about firms with a wide audience of investors.” These papers suggest that tests for advertising bias need to consider the possibility that both advertising and content respond to underlying subscriber demands. 14This identifying assumption is subject to a criticism analogous to the critique of estimating demand instrumenting with prices from other markets [Bresnahan 1997]. If certain families persistently appeal to readers of all personal finance publications in a way that is uncorrelated with observables (including the general level of print advertising), then other-publication advertising is not a valid instrument. 10