Contribution
It measured, outlet by outlet and month by month, how much front-page space four national newspapers gave to the sitting government’s corruption, and set that against how much government advertising each newspaper received that month. The comparison is within newspaper and within presidency. It is an early systematic test, for a democracy, of the transfer channel that Besley and Prat model.
Key insight
Identification comes from relative coverage. The measure compares a newspaper with itself under the same president, and with the other papers in the same month. That removes the obvious confound: friendly governments advertising in ideologically close papers that would have covered them kindly anyway. “Thus, if all papers are biased, we do not detect it with our tests.”
Assumptions
- Sample. Clarín, La Nación, Página 12 and Ámbito Financiero, 74% of national newspaper circulation, for 117 months from April 1998 to December 2007, across four presidencies (Menem, De la Rúa, Duhalde, Kirchner). There are 466 newspaper-months.
- Coverage measure. Front Pages: for each day, the share of the front page’s area covering any corruption scandal involving the current administration, measured by a research assistant with a computer, summed over the month (0 to 30). There are 254 scandals on 994 front pages. Variants are Hide (scandals another paper has reported and this one has not), Incidents (scandals only one paper covered), Affaires (covered by at least two) and Front Pages Scandal (per scandal).
- Advertising measure. Monthly federal government advertising per newspaper in millions of 2000 pesos, from the NGO Poder Ciudadano, which obtained it from the government’s Secretaría de Medios. The series starts in January 2000. For April 1998–December 1999 it was reconstructed by measuring government ad space on two random days per month and converting with the January–March 2000 overlap. “The first author was a member of the board of Poder Ciudadano during parts of our sample period.”
- Specification. OLS of Front Pages on ln(Government Advertising + 1) with newspaper and month fixed effects, then newspaper × president dummies, then newspaper × president linear trends. Newey–West standard errors.
Key results
- Table 1. The coefficient on ln(ads + 1) is −3.303 (s.e. 0.682) with newspaper and month effects, −1.652 (0.643) adding the 16 newspaper × president dummies, and −1.008 (0.589, significant at 10%) adding their trends. The authors read the halving as evidence that ideological proximity drives both advertising and coverage.
- Size. “a one standard deviation increase in government advertising (0.26 million pesos of 2000) is associated with a reduction in coverage of corruption scandals in the month by 0.31 of a cover, or 25% of a standard deviation”, evaluated at mean advertising of 0.23. Rechecked here from column 2: 1.652 × [ln(1.49) − ln(1.23)] = 0.317.
- Table 2, the channel. For 2000–2007 the association holds for advertising placed through Télam, the presidency’s news agency (−1.718, s.e. 0.679, with interactions). It vanishes for agencies that place their own advertising (0.115, s.e. 1.454).
- Table 3, leave-one-out. All eight coefficients are negative. The one excluding the Kirchner presidency, with nearly half the observations gone, is insignificant (−0.531, s.e. 1.704). The effect is largest when Clarín, the biggest paper, is dropped.
- Tables 4–5, timing. Current advertising matters more than lagged. In the Granger tables, lagged advertising predicts coverage weakly (F = 3.48 at 10% with one lag, 1.85 with two, 2.73 with three), and lagged coverage does not predict advertising (F = 0.96, 0.26, 0.08). “it is the government the one that leads”.
- Table 6, a placebo. For scandals of the police, unions, the church and piqueteros (321 scandals, 941 front pages), the coefficients are −0.373 and −0.489, both insignificant.
- Tables 7–9, scandal-level measures. Affaires carry 81% of the association (1.33 of 1.64) and Incidents 19%. Per scandal, with scandal and newspaper × scandal effects, the coefficient is −0.017 (s.e. 0.010), 2.7% of a standard deviation. Hide is +30.1 (s.e. 5.4) without the president interactions and −0.015 (s.e. 2.3) with them. The authors call this “not … robust” and conclude that advertising goes with how much space a scandal gets, not with whether or when it is first reported.
- Table 10, circulation. For Clarín and La Nación, more corruption coverage goes with higher circulation, and more Hide with lower.
- Footnote 28, an instrument. Using the government’s fiscal position times ideological proximity as an instrument for advertising gives −6.081 (s.e. 1.271) for Table 1’s column 1. This is reported in a footnote, with no first stage shown.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Within newspaper and month, and within newspaper-president pairs, more government advertising goes with less front-page coverage of the government’s corruption | strong as an association | Table 1, columns 1–2; Tables 3–4 |
| C2 | The association is specific to discretionary advertising placed centrally, and to scandals of the government | moderate | Table 2 (Télam against autonomous agencies); Table 6 (non-political scandals) |
| C3 | The government’s advertising leads and coverage follows | weak | Granger tests, with F-statistics at or near the 10% level |
| C4 | Advertising affects how much space a scandal gets, not whether it is reported | moderate | Tables 7–9; Hide is not robust |
| C5 | The negative association reflects coverage traded for transfers | weak | consistent with C1–C4 and the documented Argentine practice; an instrument in a footnote; the authors “do not offer any further interpretation” |
Concepts
- Front Pages: monthly sum of the daily share of front-page area on scandals of the current administration.
- Hide: scandals already reported by another of the four papers and not yet by this one.
- Incidents and Affaires: scandals covered by one paper, and by at least two.
- Motivated reporting: reporting “not independent of monetary transfers”, covering both bribery by the government and extortion by the paper. The negative association cannot tell these apart, but the timing favours the first.
Connections
It takes Besley and Prat (LIT-1333) as its model and argues that with non-verifiable news the government may gain from buying some outlets and not others, which their model rules out. Its identification is like Larcinese, Puglisi and Snyder’s comparison of papers’ coverage of a common event, and its private-sector counterparts are Reuter and Zitzewitz on mutual funds and Gambaro and Puglisi on Italian companies. Szeidl and Szucs (LIT-1315) cite it as “suggestive evidence” that “did not rule out plausible market-based explanations”, and they address that with private advertisers as a control.
Bearing on the record
- CASE-123, press variant. This is the documented variant the case said it was waiting for, with the scope the design allows: an evaluator (the press) whose resources (advertising) the evaluated party (the federal government) allocates at its discretion, and whose coverage of that party moves inversely with the allocation, within paper and within presidency. The case can cite THEORY-851 for the association, and should keep its own sentence from A21: “evidence of an association, not by itself proof of a particular covert intent.”
- CLAIM-404, the loop between institutions. The data show one arrow of the claim’s loop, the evaluated party’s influence on the evaluator’s resources going with the evaluator’s output. They do not show the other arrow, the evaluator’s output shaping the party’s standing. No feedback over time is estimated beyond the Granger tables.
- TERM-066. The authors measure bias relative to the other papers’ average, which is the term’s requirement of a standard set apart from the process judged, in its weakest form. If all four papers were captured alike, the test would not see it.
No instruction for machine-learning practice; nothing for the anthology.
Limitations
- An association. Fixed effects remove the stable ideological match between a paper and a president, but not month-to-month shocks that move both advertising and coverage within a pair. The instrument is reported only in a footnote.
- Advertising is one channel. Laws, licences, debt relief and access are not measured, and the authors expect substitution among them to bias the estimate toward zero. The Economist passage they quote, on Clarín’s licences and a law capping foreign stakes, is an example.
- The early advertising series is reconstructed from two days a month.
- Small units. Four newspapers, and one presidency carries much of the identifying variation (Table 3).
- The version read is not the published one. The published abstract reports 0.23 front pages and 18 percent of a standard deviation. The differences between the versions are not known here.
Open questions
- Does the published version’s smaller effect come from a changed sample or specification? A reading of the AEJ article would say.
- Would an exogenous shift in advertising, such as Szeidl and Szucs’s ownership events, move coverage in the Argentine papers too?