Does Growth Attract FDI ?

FDI is seen widely as a vital source of investment, technology transfer, and growth. The factors that attract FDI have been a longstanding source of debate. The authors present a comprehensive assessment of the accumulated evidence on one factor, the success of economic growth in attracting FDI. Meta-regression analysis is applied to 946 estimates from 140 empirical studies. The authors show that there is a robust positive correlation between growth and FDI. Significantly larger correlations are established for single country case studies than with cross-country analysis. It also appears that growth is slightly more correlated with FDI in developing countries. (Published in Special Issue Meta-Analysis in Theory and Practice) JEL O24 F21


INTRODUCTION
Although many developing economies have access to abundant natural resources, they face more limited human capital, physical capital, and technology than developed nations.
Developing countries are also constrained by corruption, the quality of their institutions, and political and economic instability. These constraints hinder capital accumulation and are a major obstacle to the efficient use of existing resources. Hence, it is not surprising that developing countries turn to international sources of economic development and economic growth, particularly foreign direct investment (FDI). 1 Developing countries seek to attract international investors by offering new and relatively unexploited markets, access to natural resources and relatively cheap labor, locational advantages, and direct and indirect incentives (Albuquerque et al. 2005;; . FDI has grown significantly in both volume and importance during the past 30 years (UNCTAD, 2012). Compared with other types of international capital flows, FDI is seen to be relatively more attractive as it offers a range of desirable characteristics to a host country. For example, it provides a relatively high degree of capital inflow stability that contributes to capital formation, and it offers the potential transfer of intangible assets such as technology, skills, management know-how, and entrepreneurship. FDI can also generate positive externalities. According to the meta-analysis by Havránek and Iršová (2010) and the survey by Clark et al. (2011), FDI is generally associated with positive spillovers, which are relatively stronger in cross-sectional and industry-level studies. 2 Another benefit of FDI is that it offers access to foreign markets (Buckley and Casson 1976;Dunning 1973;Hymer 1976;Kindleberger 1969;Vernon, 1966).
Given these desirable characteristics of FDI, it is natural that researchers and policy makers seek to identify the factors that make a host country attractive to foreign investors. One such factor is the host country's economic performance, specifically economic growth.
The focus of this paper is to identify and quantify the importance of a host country's economic growth on FDI inflows. Does economic growth attract FDI? We offer the first ever quantitative review, or meta-regression analysis (MRA), of the extant evidence (Stanley 2001).
We also assess how differences in studies (e.g. the choice of data, specification and estimation methodologies) affect the current state of knowledge on the relationship. Empirical studies are multidimensional as the data in each study differs across groups of countries, the time period, the adopted specification, and the estimation methodology. Hence, it is not clear from the existing studies whether there are similar universal FDI attracting effects, or whether effects vary by region and time period. MRA can help to dissect the literature and draw valid statistical inferences from it. The sole focus of this paper is on the effects of growth on FDI. There is a much larger literature that explores the effects of FDI on growth. This literature is not analyzed in this paper.
MRA is particularly suited to the study of the effects of economic growth on FDI (growth-FDI) literature base. Although most empirical studies find a positive relationship between economic growth and FDI, many find the opposite. For instance, the distribution of the results (the data are discussed in section 3 below) from 946 regression estimates from 140 growth-FDI studies shows that: 47 percent of the estimates are positive and statistically significant, 27 percent of the estimates are positive and statistically insignificant, 7 percent of the estimates are negative and statistically significant, and 19 percent of the estimates are negative and statistically insignificant. MRA can make sense from such apparent wide variation in results and it can explain why studies report such wide differences in the effects of economic growth on FDI. By combining the results from all comparable studies, meta-analysis increases statistical power, filters out sampling error and specification and other biases, and is thereby able to arrive at more accurate statistical inference. Moreover, MRA enables the analysis to extend beyond statistical significance and quantify the economic significance of economic growth on FDI. This effect can then be compared with other determinants of FDI, enabling policy makers to target their actions towards factors that are more effective in attracting FDI.
Our paper makes five contributions to the FDI literature. First, by necessity existing surveys provide only a selective assessment of the evidence base (e.g., Agarwal 1980;Chakrabati 2001;Lim 2001). Typically, some studies are chosen by the survey's author and a qualitative assessment is made of the literature. In contrast, we assess all 140 comparable studies, i.e., ours is a comprehensive survey of the evidence base. Second, existing reviews tend to focus on whether the effect of growth on FDI is statistically significant. However, the more interesting question is the magnitude of the effect: how large is growth's effect on FDI?
This analysis is currently missing from the literature as existing surveys do not quantify the magnitude of growth as a determinant of FDI. In contrast, our meta-analysis specifically quantifies the economic significance of growth. Third, existing surveys do not explain the heterogeneity in reported estimates. Our meta-analysis specifically maps out the distribution of reported estimates and identifies the factors that drive this heterogeneity. Fourth, none of the existing surveys explores whether the importance of growth varies by regions and/or over time.
In contrast, our meta-analysis shows that growth is significantly more important to developing countries than to developed countries. We also show that growth has a stronger effect in single country studies than in cross-country data. These findings are new to the literature. Finally, it is important to investigate time variation in order to assess whether growth is becoming more or less important as an FDI attractor over time. Some authors have suggested that growth is becoming relatively less important over time (e.g. Dunning 2002) while others challenge this view (Nunnekamp 2002). This issue remains unexplored in existing surveys. Through metaanalysis we show that the importance of growth has not diminished over time.
The paper is presented as follows. The theoretical background of the links between economic growth and FDI is presented in section 2. Section 3 discusses the data used in our meta-study, while section 4 explains the meta-regression analysis methodology. The results are presented in section 5, and the paper is concluded in section 6.

BRIEF REVIEW OF THE THEORETICAL AND EMPIRICAL LITERATURE
A broad range of potential determinants of FDI have been investigated in the literature, including the availability of an educated workforce , infrastructure (Wheeler and Mody 1992), a sound climate for international investors such as political stability , trade openness (Albuquerque et al. 2005;, comparative costs such as labor cost (Lucas 1993), taxes and tariffs Wei 2000), and access to natural resources .
When the location determinants of FDI are discussed in the theoretical literature, market size and the growth rate of host economies are treated as two of the most prominent factors . However, the net effect of economic growth on FDI is theoretically ambiguous: economic growth might have a positive effect on FDI, a negative effect on FDI, or no effect at all on FDI flows.

Economic growth as an FDI attractor
Many empirical studies find that economic growth is an incentive for FDI inflows (e.g. . There are several reasons why foreign investors might prefer faster growing markets. For example, cost efficiency of production and the realization of economies of scale and scope in production are closely linked with market size (Blonigen et al. 2007;Filippaios et al. 2003;Vernon 1966;Wang and Swain 1995). Other things equal, a growing market can be attractive to FDI because of the likelihood that a larger market will enable a more efficient scale of production through the realization of economies of scale Carstensen and Toubal, 2004). That is, growth is a measure and signal of market demand and market demand attracts FDI.  notes that while FDI location decisions depend only on recent or past earnings, they rely also on the potential and expected profitability of the specific investment project in a particular location. The prospects for market growth would need to be favorable to ensure a long-term commitment by the foreign investor.  and Zhang (2001a) argue that a higher economic growth rate, other things being equal, leads to a higher level of aggregate demand, leading to greater opportunities for making profits and, hence, increasing the incentive to invest. These incentives attract FDI to growing regions.
A higher rate of economic growth signals the size of the potential market, which could be expanded in the future. Economic growth motivates foreign firms to plan new projects or new production facilities. Regions that are experiencing rapid economic growth are also generating more profitable opportunities, and they give the promise of growing markets and growing profits.
Growing economies provide growing prospects for profitable investments. Where FDI is attracted by economic growth it will tend to be targeted at the recipient nation's domestic market rather than for exports. The size of the recipient's market can be particularly important for horizontal FDI where economies of scale are especially important. Growth, however, is unlikely to be important for vertical FDI.

Economic growth as a deterrent to FDI
Several empirical studies report negative effects of economic growth on FDI. For example, , , and  all find a significantly negative impact of economic growth in attracting FDI in developing countries.
One explanation for such empirical results is that it is a measurement artifact.  and  explain such negative associations as a result of a scaling effect; economies that grow at a faster rate than the growth in FDI will experience a decrease in FDI as a percentage of GDP.
A more causal explanation is that a recession in the host country could attract some types of FDI, especially mergers and acquisitions which can increase during a recession, as this can drive labor and capital cost downwards and thereby improve the cost structure of the firm.  and  find that while a number of industrialized countries were in recession during the early 1980s, they experienced increased FDI. In such cases, low economic growth is associated with high FDI.
A negative association between economic growth and FDI can also emerge if low economic growth means greater opportunities for future profits. For example, consider a low growth economy that is relatively capital poor but has a relatively abundant supply of cheap (underemployed or unemployed) labor and natural resources. There may here be an opportunity for FDI to profit from the relatively underutilized resources. In such cases, FDI is drawn to low growth regions in the hope of realizing unexploited opportunities for profit.

Economic growth with no links to FDI
It is entirely possible that market size and market growth might not be important considerations for export-oriented and extractive motives for FDI.  and Zhang (2001b) argue that export-oriented FDI is motivated by factor-price differentials, such as labor cost, and transportation cost from host countries to other countries in the region. For example, in Africa, extractive FDI is located in several mineral-rich countries, where market size and growth rate are not the key motivation for FDI (Akinlo, 2004). Consequently, in such cases, economic growth and FDI will be unrelated.
Hence, it is an empirical issue whether economic growth attracts, repels, or has no effect at all on FDI. It is entirely possible that growth has a positive effect on FDI in some regions, while it has a negative or even no effect in others. We apply MRA to the extant evidence to test which of these associations is supported by the data.

DATA
Like any empirical analysis, MRA requires data. In the case of MRA, this involves searching studies for relevant and comparable estimates. The basic econometric model in the empirical literature is a variant of a generic determinants of FDI model: where the variables fdi and g denote FDI and economic growth, respectively, i and t are country and time indices, x is a vector of controls, and u are the residuals (fixed country and time effects can also be included). Economic growth as a determinant of FDI requires that μ > 0.
The search and coding strategy followed the MAER-NET protocols as outlined in Stanley et al. (2013). We first commenced with a comprehensive search of the literature. We began by searching Econlit, Google Scholar, Scopus, and various other search engines. In addition to search engines, we also conducted a cited reference search on the papers that we found to have viable estimates and we also cross-referenced the references of relevant studies.
Keywords used for the search included, but was not limited to, "determinants of FDI", "drivers of FDI", "location of FDI", "market size and FDI", "economic growth and FDI", and "growth and FDI". The search for studies was terminated October 30, 2012.
The selection criteria were as follows. First, the study had to be published in a scholarly journal. We decided to exclude unpublished studies and focus only on the published literature.
Second, the study had to focus on macroeconomic relationships. Hence, studies of FDI at the firm level or a specific sectors were excluded. Third, studies that fail to report the necessary results are not included (e.g. Most and Van den Berg, 1996). This search strategy revealed 140 comparable published papers that offer regression-based estimates of the economic growth-FDI association. The studies are listed in Appendix A. These studies report a total of 946 comparable estimates of the effects of economic growth on FDI. The estimates and various characteristics of the studies were coded as variables to be used in the MRA (see below). All the coding was checked by four independent coders.
Our measure of the effect of economic growth on FDI is the partial correlation. That is, we collect estimates of the various estimates of μ (Eqn.1) and convert them into partial correlations, r. This is the correlation between economic growth and FDI, conditional on other factors that influence FDI. The partial correlation coefficient can be calculated from basic regression output as df t t r   2 , where t denotes the t-statistic of the appropriate multiple regression coefficient, and df reports the degrees of freedom. The standard error of the partial correlation is given by df r 1 2  . See Stanley and Doucouliagos (2012) for details.
As it is a correlation, care must be exercised with interpreting this measure as a causal effect. While numerous studies treat the effect of growth on FDI as a causal relation, there is also a large literature that explores the effects of FDI on growth (see ). Our below MRA shows that essentially the same inference can be drawn when we use all available estimates as when we use only those estimates that explicitly correct for endogeneity between economic growth and FDI. That is, endogeneity does not appear to be an issue in this literature.
Nevertheless, we interpret our findings as correlation and association rather than causation.
The advantage of partial correlations is that they are a standardized measure of association that is scale free and, thus, they can be meaningfully compared across the various econometric models. Unfortunately, many of the empirical studies do not provide sufficient information from which to calculate elasticities. Indeed, many studies are interested only in the direction of the effect and/or whether it is statistically significant. The partial correlation facilitates our aim to be as inclusive as possible.
The distribution of the reported estimates is illustrated in figures 1 and 2 in the form of a funnel plot, for all estimates and for estimates for developing countries only, respectively. There

THE META-REGRESSION METHODOLOGY
The analytical framework we apply is meta-regression analysis (MRA). We follow the MRA approach as developed by Stanley and Jarrell (1989), Stanley (2008), and Stanley and Doucouliagos (2012). We apply MRA to: (i) estimate the mean effect of economic growth on FDI, and (ii) identify the main factors that determine the size of the reported effect of economic growth on FDI.

(a) The mean effect of economic growth on FDI
MRA is used to estimate the mean growth-FDI effect. This "meta-average" comprises a number of dimensions studied in growth-FDI models, and it can be regarded as a reliable and statistically valid representative summary statistic of all the estimates. Note that the impact of econometric specification differences will be quantified through multiple MRA. If the metaaverage is statistically significant different from zero, then we can conclude that the literature has established that economic growth does attract FDI. The size of the meta-average then informs on how large the effect is, i.e. its economic significance and policy relevance.
The most basic approach to estimating the mean growth-FDI effect involves regressing comparable partial correlations (r) between economic growth and FDI upon a constant and an error term: where rij is the i th growth-FDI partial correlation reported in the j th study and vij is the random error. Eqn.
(2) assumes that the reported effects of economic growth on FDI vary randomly around a central effect, β0. Hence, β0 is the MRA estimate of the mean growth-FDI effect, after allowing for random sampling error. A test of H0: β0 = 0 is thus a test for whether there is a real effect from economic growth to FDI, where the magnitude of β0 informs on the size of the effect. The meta-regression of Eqn. (2) is an effective way of integrating the diverse findings from numerous models and to control for the effects of random error.
A major problem that can potentially affect any appraisal of the evidence base is the presence of publication selection bias. Selection bias arises when researchers give preference to statistically significant results, suppressing insignificant results in order to increase the probability of securing publication (Card and Krueger 1995;Stanley 2005). Publication bias can severely distort statistical inference by removing observations from the public domain (Roberts and Stanley 2005;Stanley and Doucouliagos 2012). Typically, the distortion involves inflating the meta-average, giving the appearance that the effect is greater than what it actually is. Publication selection bias is detected as a statistically significant relationship between an effect and its standard error. Absent publication bias, there should be no relationship between an estimate and its standard error. The standard test for this is to estimate the FAT-PET MRA: where SE denotes the standard error of the partial correlation (not the standard error of the regression coefficient) and ij  is the error term. 3 The Funnel Asymmetry (FAT) tests H0: βse = 0. This is a test for censoring of reported estimates (a preference for statistically significant findings). The Precision Effect Test (PET) tests H0: β0 = 0. This provides a test for the existence of a genuine empirical effect of economic growth as an attractor of FDI corrected for selection bias. Stanley (2008) points out that the PET estimate suffers from a downward bias when there is a true non-zero effect; that is, when H0:0 = 0 is rejected in Eqn.
(3) (Stanley, 2008;Stanley & Doucouliagos, 2011). This bias can be reduced by adopting a non-linear estimator that replaces SEij with SE 2 ij Doucouliagos, 2011, 2012). This model is known as precision-effect estimate with standard error (or PEESE) and involves estimating the following equation: Our meta-analysis uses only published papers as these have gone through the refereeing process. Arguably, publication bias would be more likely to be a problem when only published studies are evaluated. However, publication bias relates to selection of effects and there is no real reason to expect that unpublished studies will necessarily be less selective. Indeed, Stanley and Doucouliagos (2012) argue that the exclusion of unpublished papers (the so-called "grey literature") does not make any substantial difference to analysis of publication bias.

(b) Explaining heterogeneity in the reported effect of economic growth on FDI
Applied economics research typically exhibits excess variation (Stanley and Doucouliagos, 2012). This can be seen clearly in figures 1 and 2. The use of different datasets, different control variables, and different estimators all produce wide heterogeneity in reported estimates. MRA can be used to identify some of the key factors behind this heterogeneity. That is, we can use MRA to explain why the estimates of economic growth as an attractor of FDI differ between and within studies. This involves estimating a multiple MRA: where Z is a vector of time and regional variables and variables that reflect modeling differences. For example, Eqn. (5)  Our approach in this paper is to estimate Equations 2 to 5. We do this for all estimates available and then separately for only developing countries. There are two approaches to modeling heterogeneity in meta-analysis: the classical and the Bayesian. We follow the vast majority of meta-studies in economics and adopt the classical approach. This approach is also recommended by Stanley and Doucouliagos (2012). Applications of Bayesian approach in economics are still in the relatively minority and relatively little is known about the properties of the Bayesian approach for the MRA of economics data. For an application of the Bayesian approach see Iršová and Havránek (2013).

(c) Issues in Meta-Analysis
Data Dependence. The MRA models presented in Eqns.
(2) to (5) assume that the reported estimates, r, are statistically independent. This assumption is difficult to maintain for multiple estimates reported by the same study, which can potentially cause data dependence. 5 Our approach is to use Clustered Data Analysis, which corrects the induced lower standard errors that arise from clustering of observations within a study (Everitt et al. 2001;Hox 2002).

Study Quality.
We have tried to be as inclusive as possible in choosing studies to include in the MRA. This raises the issue of whether differences in the quality of studies might affect statistical inference. Our approach to this is to construct weighted averages, by assigning greater weight to estimates that are deemed to be of higher quality. Hence, in estimating Eqns.
(2) to (5), we do not treat each observation equally. Instead, we use precision -the inverse of the variance of a partial correlation -as weights. 6 Consequently, all models are estimated using weighted least squares. Precision is an objective measure of quality and is the standard approach in meta-analysis (Hunter amd Schmidt, 2004) and is known to produce optimal weights.  The last row of the Table 1 reports the key results using only estimates that correct for endogeneity. The results vary, but are essentially similar to those when all estimates are used.

(a) Mean Growth-FDI Effects
Below we report multiple MRA where we confirm that correcting endogeneity is not important in explaining differences in reported partial correlations. Table 1 uses all the studies, be they cross-country studies or single country case studies.  Notes: The dependent variable is the partial correlation of the effects of growth on FDI. Figures in brackets are tstatistics using standard errors robust to data clustering (clustered at the study level). Columns 1 to 3 use all estimates from all studies. Columns 4 to 6 use only estimates that relate to developing countries. Columns 1 and 4 report estimates of Eqn. (2). Columns 2 and 5 report estimates of Eqn. (3). Columns 3 and 6 report estimates of Eqn. (4). Panel B reports the results of re-estimating all models using only the subset of estimates that correct for endogeneity between growth and FDI. WLS is used for all estimations, using the inverse variance (precision squared) as the weight. We conclude from Tables 1 and 2 that when all the evidence is considered, economic growth is a statistically significant determinant of FDI. However, the size of the partial correlation is rather small. Economic growth is slightly more important for developing countries than all countries combined, but the difference is not really of practical importance.
When attention shifts to single country case studies, we find much larger partial correlations.
We cannot entirely rule out the possibility that only the more successful country case studies have been explored. Our meta-tests do not enable us to explore this proposition. Hence, we have to take the literature at face value and conclude that single country studies find a much larger role for economic growth in attracting FDI.

(b) Heterogeneity: why do reported effects vary?
In this section, MRA is applied to identify the factors that result in heterogeneity in the published results (as illustrated in figures 1 and 2). This involves estimating Eqn. (5). The variables are listed and defined in Appendix B. We commenced with a general model that included 34 explanatory variables. These results are also presented in Appendix B, columns 1 and 2 for all observations and for developing countries only, respectively. We then applied a general-to-specific modeling strategy to this general model, as recommended by Stanley and Doucouliagos (2012); statistically insignificant variables were sequentially removed. This enables greater clarity of the results. These results are presented in Table 3. Column 1 presents the results for all countries combined (all available estimates), while column 2 presents the results for developing countries only. Before discussing the results we provide a brief explanation and justification for the inclusion of the MRA variables.
Region and Data: Studies differ in the composition of the countries included in their samples.
We used the World Bank's classification to assign countries into ten regional group dummies:

Africa, Australasia, East Asia, Central and Eastern Europe, Latin America, Middle East, North
America, South East Asia, South Asia, and West Europe. 7 We use Africa as the benchmark.
These dummies are included to identify the existence of region specific growth-FDI effects.
That is, we wish to explore whether growth is more important in attracting FDI in some regions than others. This would be the case if, for example, FDI was attracted to a particular region purely because of the availability of resources, while for other regions, FDI was more motivated by growth in market demand.
In order to explore whether the reported results vary over time, we constructed the variable AveYear, which is the average year of the data used in each study. We also include Panel and SingleCountry, binary variables for whether panel data are used and whether the data relate to a single country, respectively. The benchmark here is studies that use crosssectional or time series data and a cross-country sample, respectively.

Measure:
The different measures of economic growth and FDI may be an important source of variation in empirical results. The dummy variables Gross FDI and FDI/GDP are included to explore whether measuring FDI in gross terms (total inflows) or as a ratio of GDP makes a difference to the reported results. The benchmark is all other measures of FDI, including Net FDI (FDI from foreign sources less FDI to the rest of the world) and the stock of FDI. Some studies measure growth with a lag and Growthlagged reflects these studies.
Estimator: Most of the estimates are derived from estimators that do not correct for endogeneity. The variable Endogeneity is included in the MRA in order to see whether estimates from models that correct for endogeneity are quantitatively different from those that do not. That is, the coefficient on Endogeneity informs on the size of endogeneity bias, if any. This is potentially important given the vast literature on the growth effects of FDI, which is the reverse causality of the effects of growth on FDI that we are analyzing here. We also include the binary variable Fixed in order to see if estimates that control for fixed effects differ from those that do not. An argument can also be made that studies that use Growthlagged are also correcting for potential endogeneity.

Specification:
We include 14 variables that reflect the main econometric specification differences between studies. Growth is only one of many potential determinants of growth that has been investigated by researchers. Some studies also include the level of GDP in addition to economic growth. The dummy variable Marketsize explores whether the effect that this has on the reported effects. The variables Resources, HumanCapital, DomCapital, and Infrastructure are variables that reflect host country resource and capabilities, which are also important determinants of FDI. The variables Tax rate, Labor cost, Interest rate, Tariff rate, Inflation rate, Governance, Trade and Exchange rate, reflect cost structure, competitiveness, and policy and governance outcomes in the host country, all of which can also affect FDI decisions. Finally Lagged FDI is included to capture differences between dynamic and static models.
The general-to-specific models are presented in Table 3. The negative coefficient for North America suggests that growth is slightly less important in attracting FDI than it is in Africa (or anywhere else). The statistical insignificance of AveYear means that the effects of growth on FDI have not been getting stronger or weaker over time: growth has not diminished as an important determinant of FDI. As was the case with the comparison between Tables 1 and 2, the results in Table 3 indicate that studies that focus on a single country find much larger correlations. It appears that economic growth is more highly correlated with FDI when focusing on a single country than in a pool of countries. By design, single country studies use a much smaller sample size and hence they are estimated with less precision relative to cross-country studies. However, they have the advantage that they can, in principle, offer a more nuanced analysis. Studies that use cross-country assume homogeneity between countries even though countries can differ widely.
If there is significant heterogeneity between countries, then pooling data from various countries can be problematic and unrepresentative coefficients might emerge. These concerns can be partly addressed by applying heterogeneous panel estimators. This is rarely done in this literature. MRA offers an alternative approach. By pooling the estimates from the individual case studies, we are able to control for sampling error and other differences in research design and can then identify the meta-average, or the average of the distribution of effects. Table 3 tells us that holding all other factors constant, single country case studies find, on average, much larger effects (0.14 higher for developing countries only and 0.18 higher for all estimates combined). 9 This difference is large and of practical importance.
The coefficient on Tax rate is negative. This means that studies that control for tax rates report, on average, slightly lower partial correlations than those that do not. Similarly, the coefficient on Growthlagged is also negative. This means that studies that measure the influence of current growth on FDI report larger effects than those that use a lagged value of growth. One way to interpret this is that the contemporaneous effect of economic growth on FDI is larger than the lagged effect. Another interpretation is that there could be endogeneity between FDI and economic growth. Using lagged economic growth is one way to avoid this endogeneity and doing so results in smaller effects. In this case, the MRA coefficient can be interpreted as a measure of the endogeneity bias. We also included a formal test for endogeneity with the inclusion of the Endogeneity variable. This variable is never statistically significant.
Removing Growthlagged from the MRA doesn't change the statistical insignificance of Endogeneity. This means that studies that correct for endogeneity (e.g. using IV estimation) find essentially the same results, on average, as studies that do not attempt such a correction.
de Mello (1997, 30-31) concludes that: "The association between FDI determinants and actual inflows may be stronger than that between FDI and growth such that causality may well run from growth to FDI inflows." The statistical insignificance of Endogeneity might reflect poor instrumentation strategies, so that endogeneity is not adequately controlled in the primary studies.
We explored the robustness of the results by exploring the sensitivity of key variables with respect to specification differences in the MRA. This form of sensitivity analysis is actually rare in meta-analysis. We followed a similar procedure to the one adopted by Barslund, Rand, Tarp and Chiconela (2007). Three variables were chosen as "core" variables: SingleCountry, Growthlagged, and Tax rate. That is, these three variables are included in every regression. Then, 14 other variables were included in all possible linear combinations. The WLS MRA was thus repeated a total of 16,384 times, with each MRA regression including the three core variables and various combinations of the other 14 variables. The 14 alternating variables were the publication bias variable, the regional dummies, the average year of the data, Endogeneity, and whether panel data was used. The three core variables were statistically significant 99%, 97%, and 99% in the regressions, respectively, with no instances of sign reversals. That is, they are very robust to the specification of the MRA. These robustness checks also confirm the statistical insignificance of the region dummies. The one exception is North America, which was statistically significant in 56% of the regressions. The average year of data and Endogeneity are also robust, with zero instances of statistical significance (panel data is statistically significant in only 2% of the regressions). Thus, the MRA results are robust.

SUMMARY AND CONCLUSIONS
The impact of economic growth on FDI has been a source of interest for decades. The literature contains rival theoretical predictions and much conflicting evidence. The aim of this paper is to identify the significance and the strength of the impact of economic growth in a host country on FDI inflows and to identify the impact of specification differences on the reported economic growth-FDI effects.
Our analysis is based on the available empirical evidence of 946 observations from 140 comparable empirical studies. These studies report a wide range of results, with less than half reporting a positive and statistically significant association between growth and FDI. However, by applying meta-analysis to the evidence base we are able to draw the robust conclusion that, on average, economic growth is an important determinant of FDI. MRA clearly rejects the idea that growth has no association (or even a negative association) with FDI. Growth is positively correlated with FDI in all regions, though some of the results suggest that the association is slightly weaker in North America. The correlation is slightly higher in developing countries than when all countries are combined, but the difference is not really of practical importance.
We find that economic growth plays a much more important role in attracting FDI for single country case studies than in studies that pool several countries. This difference might arise if there is significant heterogeneity between countries, and consequently analysis of crosscountry datasets understates the correlation between economic growth and FDI.
The MRA results indicate that the average partial correlation of growth on FDI is 0.18 for individual developing countries, controlling for inflation, tariffs, market size, lagged growth, and taxation. Cohen (1988)    Notes: BD means binary dummy, with a value of 1 if condition fulfilled and zero otherwise. The dependent variable is the partial correlation of growth on FDI. Figures in brackets are t-statistics using standard errors robust to data clustering. Column 1 uses all estimates from all studies. Column 2 uses only estimates relating only to developing countries. WLS is used for all estimations, using the inverse variance (precision squared) as the weight. The number of observations is reduced from 946 to 916 because of missing data for some of the moderator variables.