presented in the paper.
One of the objectives of the paper is to summarise the agoa impact reported in the selected studies. Also, to
investigate publication bias and to see if the effect is as large as reported in a couple of studies (for instance,
Collier and Venables, 2007, Nouve and Staatz, 2003). Moreover, we are also interested in whether there
is a genuine or authentic effect based on the studies selected. Additionally, does the impact depend on the
composition of countries? That is, are the studies focussing on only agoa beneficiaries, the ones reporting
larger coefficients compared to studies incorporating other non agoa countries. Finally, a number of stud-
ies on agoa have underscored the importance of apparel and textiles and reported strong impacts for agoa
beneficiaries. Using study specific variables we test whether the impact varies across product groups.
The rest of the paper is organised as follows. Section 2 introduces the data, methodology and estimation
framework used in the meta-analysis. Section 3 presents a visual guide to identifying publication bias as
well as some stylised facts of the data. The next section discusses the results while the final section, section
5 concludes the paper.
2 Data and Methodology
Search strategy
To build the database for the meta-analysis, a search was carried out at various periods between January and
June 2011. The search strategy involved querying google search, Munich personal repec archive (MPRA),
google scholar, econlit, scopus, wiley journals and jstor databases. The following key words “agoa trade
agreement”, “economic impact of agoa trade preferences ”, “African growth and opportunity act” and
“agoa trade preferences” were used in finding the studies for the meta analysis. A couple of studies were
also obtained from the references of the selected studies. The search led to 30 studies, however these were
reduced to twelve studies. Three reasons for this include: (a) some studies were working paper versions
of the published studies (all six published papers for example), (b) some studies had been published under
two or more different titles but contained the same results (examples include, Collier and Venables, 2007,
Frazer and Van Biesebreock, 2010, Nouve and Staatz, 2003, Tadesse and Fayissa, 2008), and (c) some stud-
ies even though were analysis of agoa either did not include regression analysis or looked at other aspects
of agoa. For example, Olarreaga and Ozden (2005) focussed on estimating the tariff rent in agoa apparel
while Edwards and Lawrence (2010) were interested in the impact on prices. On the contrary, Brenton and
Hoppe (2006), Lall (2005), Mattoo, et al (2003) and Paez et al (2010) were not econometric studies. This
leaves us with a sample of twelve studies consisting of 174 estimates. Of the twelve, six are published
studies.
Funnel and Galbraith plots
In investigating publication bias several authors have suggested the inspection of funnel graphs which plot
the inverse of the standard error against the effect size (partial correlation or coefficient) (Borenstein, et
al., 2009, Stanley, 2005, 2008, Stanley and Doucouliagos, 2010, Stanley, et al., 2008, Sterne, 2001). There
have been modifications and some studies show other types of funnel graphs. There is also the Galbraith
plot which shows the relationship between the t-statistic and the inverse of the standard error. When these
graphs show symmetry then it implies the absence of publication bias. However, when there are more points
to one side of the mean effect (or zero) then it is an indication of publication bias (Borenstein, et al., 2009,