On the contrary, Table (3) which presents the WLS results based on a fixed effects format indicates strongly
the presence of publication bias and also the presence of a genuine empirical effect. Column (1) reports the
basic MRA of Equation (8). The remaining columns include moderator variables. These are varied between
the columns. The final two columns, however, use the variance (σ2) of the reported coefficients as the
weights. In all seven columns, precision and the constant are significant indicating the presence of genuine
effects and publication bias respectively. With the exception of precision in column (4) and the constant in
column (3) (5% and 10% significance levels respectively) all remaining coefficients are significant at the
1% level. In two cases, the constant is positive (first two columns). Precision is also negative in two cases
(columns (2) and (4)). In the simple MRA of Equation (8), precision and the constant appear positive in all
tables presented in the main section except column (6) of Table (3) which reports a negative constant term.
Thereby indicating a positive genuine empirical effect of the agoa literature. However, including moderator
variables has caused changes in the signs of our constant term in some cases.
The number of explanatory variables in a study is positive and significant at the 1% level in all four columns
it appears. The agoa dummy (columns (3), (5) and (7)), Africa and North Africa relative to the world
(columns (4) and (5)), selection correction (columns (3), (5) and (7)), country fixed effects (columns (3)
and (7)) and cross-sectional data (columns (3) and (7)) are all significant at the 5% level and contribute to a
decrease in the reported t-statistics, all things equal. On the contrary, product fixed effects (columns (3) and
(7)), other preferences (columns (3) and (7)), logged dependent variable (columns (3) and (7)), agricultural
dummy (columns (4) and (5)), sample size (columns (2) and (7)), single country analysis (columns (3) and
(7)) and published studies (columns (5) and (7) are also significant at the 5% level but have a positive asso-
ciation with the reported t-statistics. All things equal, these variables lead to larger t-statistics.
Aggregated data (columns (4) and (5)) on the other hand is positive and significant at the 10% level imply-
ing that, holding all things equal, more aggregated definitions of exports/imports lead to larger t-statistics
relative to highly disaggregated data. Of the remaining coefficients annual data relative to monthly data
is positive and significant in column (4) while monthly data relative to quarterly data (column (5)) and
Heckman/Tobit estimators (column (4)) relative to other estimators are negatively related to the reported
t-statistics. A few of the significant coefficients are observed to reverse their signs as more moderator vari-
ables are included in the regression. Examples include, the coefficients of robust standard errors (negative
in columns (3) and (5), positive in column (7)), gravity estimators (negative in columns (3) and (7), positive
in column (5)), annual data relative to other formats (positive in (3) and (7), negative in column (5)), num-
ber of years (positive in column (2), negative in columns (3) and (7)) and number of countries (negative in
columns (3)-(5), and (7) and positive in column (2)).
13