1 Introduction
There have been several studies on the impact of the African Growth and Opportunity Act (agoa) of the
USA on Sub-Saharan African (SSA) countries. The estimates reported in these studies vary widely and
differ in terms of econometric methodology applied as well as the level of aggregation of the dependent
variable (exports and/or imports). A convenient way of summarising the coefficients reported in selected
studies is by pursuing a meta regression analysis (MRA). This is pursued in this paper using some of the
recent advances in MRA. A recent systematic review by Condon and Stern (2011) summarising the find-
ings of twenty-one econometric and non-econometric agoa studies show that (1) exports from SSA have
increased since the inception of agoa and (2) Apparel is significantly correlated with higher exports. This
paper seeks to go beyond Condon and Stern’s systematic review by performing a meta regression analysis
(MRA) on studies that estimate the impact of agoa on SSA countries.
The main contribution of this paper is extending the work of Condon and Stern (2011) to incorporate a quan-
titative summary of the agoa literature. To the best of the knowledge of the author, this is the first attempt
to investigate the agoa literature (and to some extent, trade preference literature) using a MRA approach.
Hence our contribution in this area. MRA has now become a popular way of summarising quantitative
analysis (Borenstein, et al., 2009, Stanley, 2005). There has been a phenomenal growth in its application in
several areas of economics (for example, Cipollina and Salvatici, 2010, Doucouliagos and Stanley, 2005,
2008, Feld and Heckemeyer, 2011, Rose and Stanley, 2005, among others). In this paper, our focus is
on the application of MRA towards assessing publication bias in the agoa literature. The closest study
to analysing trade preferences is Cipollina and Salvatici (2010). They apply MRA to the study of several
reciprocal trade agreements that have been ratified by the World Trade Organisation (WTO).
Several studies exist analysing the agoa preferences of the USA towards SSA countries. In spite of these
studies there are only few that make use of econometric methods to estimate the effects of agoa and this
limits the number of studies we can include in our MRA. However, the individual studies do report sev-
eral coefficients, thereby increasing our sample size. The results of the agoa studies have been mixed—
reporting varying estimates of the impact of the preference. In terms of methodology, several econometric
approaches have been undertaken. In the EU preference literature, gravity models applying Heckman selec-
tion and Poisson models tend to be very popular. However, in the agoa literature gravity modelling is less
popular. Much of the analysis are based on estimating import demand equations with one study (Seyoum,
2007) applying arima time series models. We do investigate whether these various specifications do affect
the impact measured.
The choice of studies is based on whether their emphasis is on estimating the agoa impact as well as whether
they employ econometric techniques in measuring the impact of agoa. A large number of agoa studies em-
ploy non econometric techniques in studying agoa. However, this does not limit the studies available for
performing the MRA—12 studies are used in this paper. These studies report multiple estimated coeffi-
cients varying from 1 to as many as 32 estimates and the reported impacts also vary widely. The multiple
estimates reported by the studies creates problems for estimation. One way around this problem is estimat-
ing random effects and fixed effect models to control for the within and between variation (Cipollina and
Salvatici, 2010). The fixed effects model uses the within variation while the random effects model uses a
combination of the between and within study variation. This is useful in reducing the impact of the result-
ing heterogeneity as a result of pooling various estimates. There are other approaches to get around this
problem, such as multilevel modelling—estimates are taken to be hierarchically ordered as estimates are
nested within individual studies (for example, Konstantopoulos, 2011). These are explored in the analysis