# «13TH INTERNATIONAL PUBLIC RELATIONS RESEARCH CONFERENCE “Ethical Issues for Public Relations Practice in a Multicultural World” Holiday Inn ...»

We deem newspapers as a suitable environment to test this relation. They are general oriented publication without a specific focus and -since on average they publish 200-250 articles every daythere is enough room for story selection and enough variation among different publications in editorial choices. Instead, in the case of TV and radio, the typical 25-minute long newscast includes 18-20 news, and more than half of them are in some way imposed by the real world and by the agenda setting climate of the media environment.

Moreover, within newspapers the relationship between the editorial and the sales department are weaker than within magazines, on the internet or in the professional press. From this point of view, any significant correlation between advertising and coverage would be particularly significant in a broader perspective.

** Data description**

To perform our empirical analysis, we combine four different types of data.

First, in order to gather data on newspaper coverage, we have run automatic keyword-based searches of electronic archives for an initial sample of 6 newspapers (Corriere della Sera, Repubblica, Stampa, Resto del Carlino, Mattino di Padova and Tirreno) and 13 companies: Campari, Edison, ENEL, ENI, FIAT, Finmeccanica, Geox, Indesit, Luxottica, Mediolanum, Telecom Italia, Tiscali and Tod’s. For each (company × newspaper) match, we search -on a daily basis- for the total number of articles on that newspaper containing the name of the firm. Since newspapers vary in size both cross-sectionally and in the time series, we proxy for this size by counting the daily number of articles containing the word “il” (the definite article in Italian for masculine nouns). In the empirical analysis we will focus on the daily relative frequency of articles mentioning a given company on a given newspaper, i.e. we will divide the number of articles mentioning that company by the total number of articles featured on that newspaper on that day.

Second, Nielsen kindly provided us monthly data on amount of advertising purchased by each company that is listed on the Italian stock exchange on the main Italian newspapers. Since 338 advertising expenditure refers to brands (and not to companies), we have grouped advertising data for different brands on the basis of the company owning them. The purpose of this reclassification is to match the advertising data with data on newspaper coverage, press releases and stock returns, which is at the company level. The data covers the period 2006-2007.

Third, for our sample of companies we have searched in an automatic fashion their own archives, in order to obtain information on the exact days when press releases are issued. We thus construct a press release dummy which equals one the day after a press release about a given company has been issued, and zero otherwise.

Fourth, we exploit the Yahoo! Finance website to collect data on stock quotes and transaction volume for those 13 listed companies. In particular, we use the stock quotes to compute the absolute daily return.

Summary statistics of our variables are shown in Table 1. On average the companies in our sample are mentioned on one-third of a percentage point of the total of daily articles. The distribution of this variable is strongly skewed to the left, as shown by the fact that the median number of mentions is zero percent. The distribution of monthly advertising expenditure (expressed in thousands of euros) is similarly skewed, with an average amount of about 25,000 euros and a median amount of zero. At the company level, trading volume is again positively (and strongly) skewed, with an average of around 20 million of euros and a median of 2 million. This is not the case for absolute daily return, which is only slightly skewed.

In order to gauge some sense of the heterogeneity in the data, Table 2 reports descriptive statistics at the company level. The companies issuing the largest number of press releases during the time period are ENI and Telecom, with about 200 each. On the other side of the spectrum, the most parsimonious issuers of press releases are Campari and Mediolanum, with an order of magnitude less (i.e. around 20). Regarding articles mentioning each company, FIAT and ENEL enjoy the lion’s share, with about 19,000 and 16,000 articles respectively. On the other hand, Geox is overall featured on about 300 articles, while Campari appears on around 500 articles. FIAT and ENEL are characterised by the highest ratio between articles featuring them and number of press releases issued by them, while Geox and Finmeccanica have the lowest ratio.

For each company we also report the mean relative frequency of articles mentioning that company over the total number of articles being published by each newspaper (column 5). We can also compute this relative frequency conditionally on the presence or the lack of a press release being issued by that company the day before (columns 6 and 7). We can then calculate the percentage change in the relative frequency of articles in press-release vs. non-press-release days (column 8). In a nutshell, the average relative frequency of articles about a company in the lack of a press release is informative about the newsworthiness of that company when the company itself does not produce any additional news. On the other hand, the percentage change in the relative frequency of articles in the presence of a press release would be indicative of the capacity of each company to create additional media coverage.

With a cursory look at the table one can see that the largest companies in our sample, i.e. Fiat, Enel, Eni and Telecom Italia, do obtain the largest amount of newspaper coverage in the lack of an immediately preceding press release. On the other hand, smaller companies like Campari, Geox and Mediolanum, which start with a low level of coverage in the lack of a press release, enjoy the largest increase in newspaper coverage after the issuance of a press release. Quantitatively speaking, the average change is more than threefold for Campari, and more than twofold for Geox and Mediolanum.

Table 3 displays descriptive statistics at the newspaper level. Overall there are about 57,000 articles mentioning our sampled companies. In relative terms, Stampa is the outlet dedicating more room to companies, while Resto del Carlino and Tirreno dedicate the least. Similarly to what done in Table 2, for each newspaper we compute the relative frequency of articles mentioning one of our sampled companies, respectively in the presence and in the lack of a press release being issued the 339 previous day. We can also calculate the percentage change in coverage when moving from a nonpress-release to a press-release day. From this point of view, it turns out that Corriere della Sera and Stampa are the outlets with the largest average increase in coverage after a press release.

** Results**

As mentioned in the introduction, we are especially interested in the relationship between media coverage of companies and advertising expenditure, controlling for potentially confounding factors.

To get a first glance at the correlations in the data, we first compute monthly (instead of daily) relative frequencies of stories about a given company on each newspaper: We are thus left with 1872 observations at the (company × newspaper × month) level. Second, we regress those relative frequencies against a set of fixed effects for each company and each newspaper, plus dummies for those cases where the company owns a significant stake in the newspaper itself. This is true for the match between FIAT and Stampa, and for the one between Corriere and FIAT, Telecom Italia and Tod’s. Finally, we compute the residuals of the estimated regression. We do the same (i.e. regress it against a set of fixed effects and obtain residuals) for the total amount of ads being purchased by each company on each newspaper the month before. Figure 1 displays a scatter plot of the coverage residuals against the ads residuals, together with the corresponding linear fit. The relationship is positive and strongly significant. 6 Controlling for ownership links and time-invariant features of each company and each newspaper, our data suggests that companies buying more ads on a given newspaper obtain significantly more coverage on that newspaper.

In order to delve further into this correlation, we run a set of fixed effects regressions with the relative frequency of articles mentioning company c on newspaper n on day t as the dependent variable. As mentioned in the introduction, we first focus on advertising expenditure and the issuance

**of press releases by each company. More formally, we run the following type of regression:**

y nct = α n + β c + γ ⋅ ADS nc,t −1 + ζ ⋅ pr _ d c,t −1 + φ ⋅ ADS nc,t −1 × pr _ d c,t −1 + ε nct (1) where y nct is the relative frequency of articles mentioning company c appearing on newspaper n on day t, α n and β c are respectively a newspaper and a company fixed-effect, ADS nc,t −1 is the monetary amount of ads being purchased by company c on newspaper n the month before, pr _ d c,t −1 is a dummy which equals one if company c issued a press release on day t-1, and ε nct is the error term. In order to properly take into account the fact that the error term might be serially correlated within company-newspaper pairs (even after controlling for company and newspaper fixed effects) and hence overestimate the precision of our results, we correspondingly cluster the standard errors at the (company × newspaper) level. 7 Our regression output is displayed in Table 3, whereas we proceed by expanding the set of explanatory variables. Thus in column [1] we simply control for purchased ads, we then add the press release dummy in column [2] and the interaction between this dummy and ads in column [3].

In column [4] we add the two ownership dummies for Corriere and Stampa we have mentioned above.

Across all specifications advertising expenditure is positively and significantly correlated with media coverage. The effect is actually smaller in size when controlling for ownership links (column [4]). In terms of magnitudes, a coefficient of 0.002 in columns [1]-[3] implies that an additional expenditure 6 Standard errors are clustered at the (company × newspaper level), in order to account for withincluster correlation in the error term.

7 See Bertrand et al. (2004).

340 of 50,000 euros per month by a given company (somewhat less than a standard deviation) is associated with one additional article every one thousand about that company. Since on average there are around 13,000 total articles per month, this correlation translates in an increase of about 13 articles per month.

The issuance of a press release is a very significant (and positive) predictor of newspaper coverage, across all specifications. When not controlling for the interaction between ads and press releases (i.e.

in column [2]), an additional press release is associated with around one and a half additional article mentioning that company every one thousand.

Column [3] shows that the interaction between press releases and advertising is positive and significant at the 5% confidence level. The effect is more strongly significant (and still positive) when controlling for the ownership dummies (column [4]). The magnitude of the conditional effect can be calculated as follows: at the mean level of monthly ads (about 25,000 euros per month) an additional press release is associated with 1.375 additional articles every one thousand (0.12 +

0.0007 x 25). On average, if a firm doubles the amount of advertising, the effect rises to 1.55 additional articles every one thousand. It would jump to around 2 additional articles every one thousand when considering a standard-deviation increase (i.e., around 66,000 euros per month).

The discrepancy in the size and significance of those findings when controlling or not controlling for the ownership dummies is consistent with the fact that companies holding a stake in a given newspaper are not constrained by the issuance of press releases in affecting media coverage about themselves.

In columns [5]-[9] we proxy “residual” real world events pertaining to a company (i.e. those not captured by the issuance of a press release) by controlling for rc,t −1, the absolute daily return of stock c on day t − 1, by itself and interacted with advertising expenditure. 8 It must be noticed that sample size decreases, since we are only considering days that are immediately preceded by a trading day, i.e. we do not include Sundays and Mondays in the sample.

We find that larger absolute returns are significantly correlated with wider media coverage: in column [5] a coefficient of about 3 implies that a one percentage point increase in the absolute daily return is associated with an increase of 0.03 percentage points in the amount of coverage.

The coefficient on the interaction term between monthly ad expenditure and the absolute stock return would be informative about whether newspaper coverage of a given company differentially reacts to the same stock return as a function of the amount of ads purchased by that company on that newspaper. In column [6] we find that this coefficient is positive and significant at ordinary confidence level. To get a sense of the magnitude of the estimated effect, consider a one standard deviation increase in the amount of ad purchases: the reactivity of newspaper coverage to the absolute stock return would jump from 2.151 to 3.67, i.e. it would increase by more than 70 percent.

However, the interaction term is not statistically significant in column [7], i.e. when controlling for the ownership dummies. One should also notice that the significance and magnitude of the ads variable, the press release dummy and the interaction term between the latter is practically unaltered when controlling for past absolute return.

A relevant concern here is that the estimated reactivity of newspaper coverage to the absolute stock return -by itself and interacted with ads purchases- does disguise different correlations in the case of positive vs. negative returns. In columns [8] and [9] we address this issue by separately considering positive and negative returns, properly interacted with lagged ads purchases. In fact, both