Table 3 MAIVE estimates tend to be closer to 0 than PET-PEESE in meta-analyses

From: Spurious precision in meta-analysis of observational research

 

All effect sizes

PET-PEESE significant

 

Absolute

%

Absolute

%

(a) All meta-analyses

    

∣MAIVE∣ > ∣PET-PEESE∣

111

35.8

63

29.9

∣MAIVE∣ < ∣PET-PEESE∣

199

64.2

148

70.1

Total

310

100

193

100

(b) Meta-analyses with F > 10

    

∣MAIVE∣ > ∣PET-PEESE∣

87

32.6

44

24.6

∣MAIVE∣ < ∣PET-PEESE∣

180

67.4

135

75.4

Total

267

100

172

100

(c) Meta-analyses with F > 100

    

∣MAIVE∣ > ∣PET-PEESE∣

70

29.2

36

22.4

∣MAIVE∣ < ∣PET-PEESE∣

169

70.7

125

77.6

Total

239

100

151

100

  1. Notes: The table compares the results of MAIVE and PET-PEESE for the sample of economics meta-analyses56. The table separates the cases in which the estimated underlying effect in PET-PEESE is statistically significant at the 5% level (right) and when all PET-PEESE estimates are considered (left). In both cases, MAIVE estimates are typically closer to zero (that is, smaller in absolute value) than PET-PEESE estimates. The difference is larger for statistically significant effects and for meta-analyses with a large F-statistic in the first-stage regression of MAIVE; the F-statistic therefore measures the strength of the MAIVE instrument. The larger the F-statistic, the more reliable MAIVE is, althought the difference in performance is not large. 
Back to article page