In a previous post, I highlighted the conceptual framework used to analyze the link between statewide natural resources and educational attainment in a conference paper I wrote with a collegue. You can view that post here.

In this post, I will discuss the methodology, findings and analysis.

**Methodology**

The selection and collection of various state-level education and natural resource dependency variables required the use of several public data sources. To collect variables on natural resource dependency in each of the 50 states three government databases were utilized- The U.S. Energy Information Administration (EIA) State Energy Data System (SEDS), the Bureau of Economic Analysis (BEA) Regional Data Database, and the U.S. Census Bureau’s North American Industry Classification System (NAICS). To collect variables on broad-based educational attainment in each of the 50 states statistics from The National Center for Public Policy and Higher Education’s *Measuring Up 2008* report and data from the National Center for Higher Education Management Systems (NCHEMS) were utilized. A correlation analysis using Pearson’s Correlation Coefficient, Chi-Square Test Statistics and multiple regressiona analysis were conducted.

**Findings and Analysis**

*Correlation Analysis*

Our analysis began with a study of the bivariate correlations between the state-level indicators of natural resource dependency and educational attainment variables using the Pearson product-moment correlation (r). This analysis yielded several significant results, which are reported in Table 1 below.

While several significant relationships were found, the strongest relationships were found to be between the percentage of state GDP derived from mining activities (PCTGDP), the percentage of state tax revenue derived from mining activities (PCTTAX) and the percentage of 18-24 year olds enrolled in postsecondary education (YNGADLTENR), and state/local support for higher education operating expenses per capita (SUPPCAP).

A Pearson product-moment correlation coefficient was computed to assess the relationship between the percentage of state tax revenue derived from mining activities and the percentage of 18-24 year olds enrolled in postsecondary education .There was a significant negative correlation between the two variables, r (48) = -.417, p= .003. A Pearson product-moment correlation coefficient was also computed to assess the relationship between the percentage of tax revenue derived from mining activities and state/local support for higher education operating expenses per capita. There was a significant positive correlation between the two variables, r (48) = .641, p= .000.

The first finding (the negative correlation between percentage of tax revenue derived from mining activities and the percentage of adults enrolled in postsecondary education) is consistent with the concepts derived from our conceptual framework. As discussed earlier, a young person incurs the costs of education only if he or she expects to be employed in the non-natural resource sector. Thus, the sectoral composition of the economy serves as a signal to citizens. States which derive a disproportionate share of their tax revenue from the mining industries are likely to have large and well-established industries and employment in the mining sector. The second finding (the positive correlation between the percentage of tax revenue derived from mining activities and state/local support for higher education per capita), however, is not consistent with the concepts derived from our conceptual framework. The literature which formed the basis of our conceptual framework posited that the rent seeking behavior associated with natural resource exploitation can provide governments that receive a disproportionate share of their revenue from natural resource rents with disincentives for investing in an education system that will pull workers from the natural resource sector and push them in to non-natural resource sectors of employment.

*Chi-Square Test Statistics*

To analyze the relationship between natural resource dependency and broad educational attainment metrics, we conducted Chi-square test statistics using the variables GDPQTL, PREPSCR, PARTSCR and BNFTSCR. The independent variable GDPQTL, which we used as an index of state dependency on natural resources, was created by recoding the percentage of state GDP derived from mining activities (PCTGDP) into quartiles. The dependent variables PREPSCR, PARTSCR and BNFTSCR are the letter grades assigned to each state in the *Measuring Up, 2008 *report for preparation of high school students for postsecondary education (PREPSCR), participation in post-secondary education (PARTSCR) and the level of benefits to the state realized through postsecondary credential attainment (BNFTSCR). The crosstabs and test statistics held significant findings in the PREPSCR and BNFTSCR analyses. No significant findings were yielded from the PARTSCR analysis.

As illustrated in Table 2 above, 66.7% of states receiving an “A” for preparation were in the lowest quartile of natural resource dependency and approximately 80% of states receiving a “D” for preparation were in the two highest quartiles of natural resource dependency. The Linear-by-Linear Association test statistic was 5.819 and the probability was 0.016.

As illustrated in Table 3 above, approximately 80% of states receiving an “A” for benefits were in the lowest quartile of natural resource dependency. The Linear-by-Linear Association test statistic was 12.310 and the probability was .000. This finding is consistent with our conceptual framework because it shows that in states with low natural resource dependency incentives exist for the attainment of a postsecondary credential.

*Multiple Regression Analysis*

The findings above, along with concepts gained through the formulation of our conceptual framework, informed a series of multiple regressions.

In order to gauge whether or not the relationship between the percentage of young adults enrolled in postsecondary education and the percentage of tax revenue derived from mining activities observed in the correlations matrix remained robust in the presence of other independent variables, a regression model using YNGADLTENR as the dependent variable was conducted. This regression yielded the following results.

As illustrated in Table 4b above, the model is statistically significant (p < 0.05). The R-squared, found in Table 4a, is 0.297, meaning that approximately 30% of the variability in YNGADLTENR is accounted for by the independent variables in the model. However, the adjusted R-squared indicates that only 18% of the variability in YNGADLTENR is accounted for by the model after taking into account the number of predictor variables in the model. Table 4c illustrates that no single variable contributed significantly to the regression model and that, despite our bivariate correlation findings, PCTTAX does not significantly contribute to the model once other variables are taken in to account.

The correlation analysis yielded rather weak but significant negative correlations between the percentage of the adult population with an associate’s degree or higher (ADASSOC) and the percentage of the state’s GDP derived from mining activities (PCTGP) and the number of residents employed in mining (MINEMP). In order to gauge the strength of these relationships in the presence of other independent variables, a regression model using ADASSOC as the dependent variable was conducted. This regression yielded the following results.

As illustrated in Table 5b above, the model is statistically significant (p < 0.05). The R-squared, found in Table 5a, is 0.336, meaning that approximately 34% of the variability in ADASSOC is accounted for by the independent variables in the model. The adjusted R-squared indicates that approximately 23% of the variability in ADASSOC is accounted for by the model after taking into account the number of predictor variables in the model. Table 5c illustrates that only MINEMP had significant negative regression weight after controlling for other variables in the model.

Another regression model, using the percentage of the adult population with a bachelor’s degree or higher (ADBCHLR) as the dependent variable, yielded similar, although not as robust, findings as the regression model using ADASSOC. The results from that regression are below.

As illustrated in Table 6b above, the model is statistically significant (p < 0.05), although not as robust as the ADASSOC model. The R-squared, found in Table 6a, is 0.293, meaning that approximately 29% of the variability in ADBCHLR is accounted for by the independent variables in the model. The adjusted R-squared indicates that approximately 18% of the variability in ADBCHLR is accounted for by the model after taking into account the number of predictor variables in the model. Table 6c illustrates that, like the ADASSOC model, only MINEMP had significant negative regression weight after controlling for other variables in the model.

Lastly, to gauge whether or not the relationship between SUPPCAP and PCTTAX observed in the correlations matrix remained robust in the presence of other independent variables, a regression model using SUPPCAP as the dependent variable was conducted. This regression yielded the following results.

As illustrated in Table 7b above, the model is statistically significant (p < 0.01). The R-squared, found in Table 7a, is 0.483, meaning that approximately 48% of the variability in SUPPCAP is accounted for by the independent variables in the model. The adjusted R-squared indicates that only 40% of the variability in SUPPCAP is accounted for by the model after taking into account the number of predictor variables in the model. Table 7c illustrates that no single variable contributed significantly to the regression model and that, despite our bivariate correlation findings, PCTTAX does not significantly contribute to the model once other variables are taken in to account.

In the next (and final) post concerning this conference paper, I will discuss the implications of the findings highlighted above as well as ideas for further research.

Pingback: Conference Paper- Pass the Books. Hold the Oil: Is There a Link Between Natural Resources and Educational Attainment- Implications and Ideas for Further Research | higheredmaven