QUESTION 2 - ELECTION FRAUD EXPERIMENT In a paper titled "Institutional Corruption and Election Fraud: Evidence from a Field Experiment in Afghanistan", Callen and Long (2015) study the impacts of a new monitoring technology on the manipulation of vote totals during the 2010 parliamentary elections in Afghanistan. Specifically, they test whether announcing the use of this technology at a given polling center to election officials reduces fraud. In the run up to elections, they deliver a letter explaining that the monitoring technology will be used at that location to a randomly selected set of polling center managers in 238 polling centers from an experimental sample of 471 polling centers. Table 9 examines the impacts of sending the letter on the votes received by the candidate with the strongest political connections. The data from four regressions for polling centers (PCs) are reproduced here, with standard errors in parentheses: TABLE 9 SPATIAL TREATMENT EXTERNALITIES Votes for the most connected candidate (1) (2) (3) (4) Letter treatment (= 1) -4.080** -4.183** -4.290** (2.009) (1.982) (1.956) -4.159** (1.980) Any PCs treated within 1 km (= 1) -6.877* -6.742* (3.512) (3.486) Total PCs within 1 km -0.597 -0.499 -1.256 (0.566) (0.564) (0.806) Any PCs treated within 1-2 km (= 1) -4.738 -4.681 (4.244) (4.240) Total PCs within 1-2 km 0.103 0.223 (0.378) (0.392) 1 treated PC within 1 km (= 1) -6.457* (3.613) 2 treated PCs within 1 km (= 1) -5.831 (3.882) 3 treated PCs within 1 km (= 1) -3.007 (4.858) 1.459 4 treated PCs within 1 km (= 1) 5 treated PCs within 1 km (= 1) Constant (5.620) -1.334 (6.929) 28.064*** (6.017) 30.543*** (6.043) 32.378*** 32.697*** (7.004) (6.987) R 0.276 0.290 0.292 0.294 Trimming top 1 percent of votes for interacted candidate type Number polling centers Yes Yes Yes Yes 439 439 439 439 Number candidate-polling substation observations 1,841 1,841 1,841 1,841 Mean dep. var. control + no treated PCs 0-2 km 42.939 42.939 42.939 42.939 (a) Explain, briefly but precisely, the meaning of the coefficient estimate of -4.080 on Letter treatment (=1) in regression (1) in the context of this evaluation. (b) Interpret the coefficient estimate of -6.877 on Any PCs treated within 1km (=1) [a dummy variable for whether any other centers within 1km of a given polling center received the letter treatment] in regression (2) in the context of this evaluation. What does it tell us in terms of spillovers/externalities? (c) What does your answer to part (b) imply about whether the coefficient estimate from part (a) is an underestimate or overestimate (or neither) of the total impact of the letter treatment?