State fragility as a multi-dimensional construct for international entrepreneurship research and practice

Abstract

While the international entrepreneurship literature has begun to expand its scope to include the world’s poorest and least-developed countries, it has not yet reached a consensus about how to conceptualize and measure the relevant cross-national differences that shape the wide range of entrepreneurial activities across the developing world. To address this gap, we explore the potential contribution of the multi-dimensional construct of “state fragility,” as developed in related fields, as an orienting framework for developing-country entrepreneurship research. Instead of viewing all developing country states as universally weak, the state fragility construct provides a conceptual framework to identify which types of capabilities are weak, to which degree, and in what configuration, thus advancing efforts to systematically compare and contrast the wide diversity of entrepreneurial processes and outcomes found across the developing world.

Appendix: Confirmatory Factor Analysis of State Fragility Indicators

To identify the variables for our confirmatory factor analysis of the indicators that make up the latent “state fragility” construct, we included 22 publicly-available indicators used in Rice and Patrick’s (2008) “Index of State Weakness” and Marshal and Cole’s (2016) “Polity IV State Fragility Index.”³ For accuracy, we pulled these measures from their original sources, rather than from the existing state fragility indexes. For each measure, we used 2016 data, which was the last year available in every dataset. We tested the model using the “sem” command in Stata.

In our initial analysis, we loaded each variable onto the same factor hypothesized in the existing state fragility indexes listed above. This preliminary analysis led to a poor overall model fit, as some variables loaded weakly along the predicted dimensions. For instance, “regulatory quality” from the Kaufmann et al. (2016) index proved to be more highly correlated with the “Political” factor than with the “Economic” factor, where Rice and Patrick (2008) originally placed it. We therefore included this indicator with the political, rather than the economic, factor. We also found that “Voice and Accountability,” “Electoral Freedom,” and the “Democracy Index” did not load onto the “political” factor as theorized, so we removed these variables. As illustrated in Table 4 that reports our results, this established the presence of a strong political factor that included rule of law, control of corruption, government effectiveness, and regulatory quality from the Kaufmann database. This empirical finding is consistent with state fragility researchers that identify “regulatory capacity” as a separate characteristic of government than regime type (Gisselquist, 2014; Grävingholt et al., 2015; Hanson & Sigman, 2013).⁴

Table 4 Best fit results of confirmatory factor analysis of state fragility indicators

We also found that none of the “economic” variables included in existing state fragility indexes loaded on a clearly defined economic factor. When we removed these measures from the model, overall fit improved, thus further casting doubt on the presence of a distinct “economic” dimension within the data. We therefore removed it as an independent factor in our analysis; a decision consistent with a broader approach in indexes of state capabilities that view economic variables as an outcome of state capacity rather than as a constituent dimension (Hanson & Sigman, 2013). Child mortality, access to improved water, life expectancy, and human development all loaded on the “Social-Welfare” factor, as hypothesized, but primary school completion rate and undernourishment did not and were dropped. Similarly, political stability and political terror loaded on the “Security” dimension, but “major episode of political violence,” and “coup d’états” did not and therefore were not included in our final model. In total, we found that 10 variables loaded on three factors at .999 confidence levels.

We present the standardized coefficients, p values, covariance between each first-order factor, and fit statistics (X ², RMSEA, CFI, TLI, SRMR) of the best-fit final model in Table 4. While X ² and RMSEA are typically reported in confirmatory factor analysis, they are sensitive to sample size and are thus often not interpreted (Ault, 2016; Busenitz et al., 2000). All remaining fit statistics suggested a strong fit of the final model.

We next tested the discriminant validity of the three factors found in the model. As shown in the table, the Political and Social-Welfare dimensions appear to be correlated. We thus tested whether these were distinct factors by loading all 8 of these variables onto a single factor. CFI, TLI, and SRMR all deteriorated past conventional cutoffs for acceptable fit (Busenitz et al., 2000). Moreover, the standardized factor loadings for nearly all variables previously included on the “Social-Welfare” dimension fell below .80, thus suggesting that the first-order factors are related, but distinct.

Having completed the factor analysis, our final task was to construct a state fragility index based on the standardized factor loadings. To develop final rankings, we first inverted the scores for all variables, except child mortality and political terror, so that a higher value would indicate greater fragility. We then standardized the scores and combined them across each factor, weighted by the factor loadings. We again standardized the combined scores to make three comparable first-order factors. To create a combined state fragility score for each country, we took a straight average of the three indicators. Table 5 shows the overall state fragility scores and each indicator for the 156 countries in our sample. The full index is also available in Microsoft Excel format at: http://www.joshuakault.com/ault-spicer-state-fragility-index.

Table 5 State fragility scores for 156 countries: factor results