4.2.3 Infrastructure & Education
The next test will be an analysis the relationship between the remaining predictors and the outcome variables. But at first we shall eyeball the data and distributions for a while. The lowest aggregate score on the infrastructure index reached was 1 (by e.g. Telorman), the highest score was achieved by Bucharest (10). A simple regression indicates a considerable effect and fits the data adequately (R² = .6215, cf. Figure 4.3).
An interesting observation is that the ranking order of the Transylvanian counties becomes clearer. Cluj (CJ), Timișoara (TM) Arad (AR) or Sibiu (SB) exceed Hunedoara (HD), Harghita and CarașSeverin (CS) not only with regard to per Capita GDP but also concerning infrastructure. Constanța (CT) on the other hand demonstrates that infrastructural well endowed counties must not be worse off just for belonging to the East of Romania, so do Bacau (BC), Suceava (SV) and Iasi (IS),[1] even though these better endowed counties in the East nevertheless seem to suffer from infrastructural shortages of their neighbor counties which of course have to be traversed when it comes to logistics (cf. Figure 4.4). For instance, travelling from Galați to checkpoint Nădlac (Arad) on European (and national) Roads would require the indirect route via Brăila, Ialomița, Bucharest and so on.
GDP per capita in Romania follows a similar pattern on the map (cf. Figure 4.5). But the maps and the distribution in Figure 4.3 also clarify that infrastructure does of course not account for all the variance. With regard to FDI the slight nonlinear relation between GDP per capita and foreign economic activity – as observed in chapter 3.5 – becomes a real cubic function once Yi is replaced by IAi as predictor variable (cf. Figure 4.6). The effect can nonetheless be tracked sufficiently by linear regression, even though the linear model underestimates the role of infrastructure. The same is true for overall economic activity, TEAi, which follows the same pattern as FDI, though foreign firms seem to react even more sensitive to infrastructure than domestic firms.
The effect of human capital, Hi, can likewise be tracked well via linear models. Again, a simple regression analysis indicates a relatively strong effect and a well fit of the data, only diminished by one major outlier (county Ilfov without students but a very high per capita income). This outlier actually is none as students from Ilfov are obviously enrolled in Bucharest. For illustrational purposes, both Ilfov and Bucharest were removed from Figure 4.7.[3]
After the removal of Ilfov and Bucharest three more outliers draw attention. These are Dolj (DJ), Galați (GL) and Iași (IS). All three perform better than their neighbors but worse than counties with a better infrastructural endowment or a better location. The relation between FDI and human capital (R² = .76) suffers from many outliers, which become especially visible if Bucharest and Ilfov are removed from the sample (R² = .35). But on the other hand, the relation is stronger and relatively linear with regard to total economic activity (R² = .642). Both predictor variables will be used now for multiple linear regressions in the form of
Yi= α + β_{1}IA_{i} + β_{2}H_{i} + e
A detailed overview on the results for GDP per capita (Yi) is presented in Table 4.4 to Table 4.8. The results suggest a strong and highly significant impact of both predictors while the data is fitted well.
Table 4.4 : Descriptive Statistics (SPSS) – GDP (Romania)  
Variable 
Mean 
Std. Deviation 
N 
Yi (GDP per Capita) 
11,871.14 
4,219.73 
42 
IAi (Infrastructure and International Accessibility) 
3.92 
2.35 
42 
Hi (Human Capital or Availability of Qualified Labor) 
17,058.66 
40,631.11 
42 
Source: Own table, own calculations 
Table 4.5: Correlations (SPSS) – GDP (Romania)  
Yi 
IAi 
Hi 

Pearson Correlation  Yi 
1.000 
.789 
.764 
IAi 
.789 
1.000 
.588 

Hi 
.764 
.588 
1.000 

Sig. (1tailed)  Yi 
. 
.000 
.000 
IAi 
.000 
. 
.000 

Hi 
.000 
.000 
. 

Source: Own table, own calculations 
Table 4.6: Model Summary (b) (SPSS) – GDP (Romania) 


Model 
R 
R Square 
Adjusted R Square 
Std. Error of Estimate 
Change Statistics 
DurbinWatson 

R Square Change 
F Change 
df1 
df2 
Sig. F Change 
2.117 

1 
.872(a) 
.760 
.747 
2,121.41 
0.760 
61.610 
2 
39 
.000 

a. Predictors: (Constant), Hi, IAi;b. Dependent Variable: Yi  
Source: Own table, own calculations 
Table 4.7: ANOVA (SPSS) (b) – GDP (Romania)  
Model 
Sum of Squares 
df 
Mean Square 
F 
Sig. 

1  Regression 
554,536,196.94 
2 
277,268,098.47 
61.610 
.000(a) 
Residual 
175,514,798.65 
39 
4,500,379.45 

Total 
730,050,995.59 
41 

a. Predictors: (Constant), Hi, IAi ; b. Dependent Variable: Yi  
Source: Own table, own calculations 
Table 4.8 : Coefficients (SPSS) (a) – GDP (Romania)  
Model 
Unstandardized 
Standardized 
t 
Sig. 
95 % Confidence 

B 
Std. Error 
Beta 
Lower B.  Upper B.  
1  (Constant) 
7,408.715 
682.431 
10.856 
.000 
6,028.38 
8,789.06  
IAi 
931.650 
174.552 
.518 
5.337 
.000 
578.585 
1,284.716 

Hi 
.048 
.010 
.459 
4.728 
.000 
.027 
.068 

a. Dependent Variable: Yi  
Source: own table, own calculations 
Both independent variables correlate highly (infrastructure r = .789; qualified labor r = .764) with economic performance, captured here as GDP per capita by county at a highly significant level (> 99.9 %). With an R² = 0.76 the model fits the data well and the constant α of RON 7,408.72 is by far smaller than average GDP per capita on county level (RON 11,871.14). The standardized coefficients for infrastructure of .518 and for human capital of .459 indicate both a strong, positive influence of the predictor variables on economic performance (GDP per capita). All parts of the model are also individual highly significant at a > 99.9 % level. These results are in line with frequent complaints in the business press about the difficult connection to the European business network and shortages on (qualified) labor markets. Hence, the impact of the two predictors on FDI and total economic activity remains to be tested next.
Table 4.9: Descriptive Statistics (SPSS) – FDI (Romania)  
Variable 
Mean 
Std. Deviation 
N 
FDIi (Foreign Firms per capita) 
0.0045 
0.0062 
42 
IAi (Infrastructure and International Accessibility) 
3.92 
2.35 
42 
Hi (Human Capital or Availability of Qualified Labor) 
17,058.66 
40,631.11 
42 
Source: Own table, own calculations 
Table 4.10: Correlations (SPSS) – FDI (Romania)  
FDIi 
IAi 
Hi 

Pearson Correlation  FDIi 
1.000 
.710 
.873 
IAi 
.710 
1.000 
.588 

Hi 
.873 
.588 
1.000 

Sig. (1tailed)  FDIi 
. 
.000 
.000 
IAi 
.000 
. 
.000 

Hi 
.000 
.000 
. 

Source: Own table, own calculations 
Table 4.11: Model Summary (b) (SPSS) – FDI (Romania) 


Model 
R 
R Square 
Adjusted R Square 
Std. Error of Estimate 
Change Statistics 
DurbinWatson 

R Square Change 
F Change 
df1 
df2 
Sig. F Change 
1.801 

1 
.906(a) 
.821 
.812 
0.0027 
0.821 
89.589 
2 
39 
.000 

a. Predictors: (Constant), Hi, IAi ; b. Dependent Variable: FDIi  
Source: Own table, own calculations 
Table 4.12: ANOVA (SPSS) (b) – FDI (Romania)  
Model 
Sum of Squares 
df 
Mean Square 
F 
Sig. 

1  Regression 
.001 
2 
.001 
89.589 
.000(a) 
Residual 
.000 
39 
.000 

Total 
.002 
41 

a. Predictors: (Constant), Hi, IAi ; b. Dependent Variable: FDIi  
Source: Own table, own calculations 
Table 4.13: Coefficients (SPSS) (a) – FDI (Romania)  
Model 
Unstandardized 
Standardized 
t 
Sig. 
95 % Confidence 

B 
Std. Error 
Beta 
Lower B.  Upper B.  
1  (Constant) 
.000 
.001 
–.489 
0.628 
–.002 
.001  
IAi 
.001 
.000 
.30 
3.583 
0.001 
.000 
.001 

Hi 
1.06E007 
.000 
.697 
8.319 
0.000 
.000 
.000 

a. Dependent Variable: FDIi  
Source: own table, own calculations 
Again, the both predictors do a good job in tracking the location decision of foreign investors. The linear model underestimates the role of infrastructure (β = .3), which does not follow a linear trend but actually appears as a cubic function in the plot. Nonetheless, the explained variance (R² = .821) indicates a well fit of the data and the individual significance levels are highly satisfying (>99.9). The results are similar for total economic activity.
Table 4.14: Descriptive Statistics (SPSS) – TEA (Romania)  
Variable 
Mean 
Std. Deviation 
N 
TEAi (Total Economic Activity) 
.071 
.022 
42 
IAi (Infrastructure and International Accessibility) 
3.92 
2.35 
42 
Hi (Human Capital or Availability of Qualified Labor) 
17,058.66 
40,631.11 
42 
Source: Own table, own calculations 
Table 4.15: Correlations (SPSS) – TEA (Romania)  
TEAi 
IAi 
Hi 

Pearson Correlation  TEAi 
1.000 
.764 
.801 
IAi 
.764 
1.000 
.588 

Hi 
.801 
.588 
1.000 

Sig. (1tailed)  TEAi 
. 
.000 
.000 
IAi 
.000 
. 
.000 

Hi 
.000 
.000 
. 

Source: Own table, own calculations 
Table 4.16: Model Summary (b) (SPSS) – TEA (Romania) 


Model 
R 
R Square 
Adjusted R Square 
Std. Error of Estimate 
Change Statistics 
DurbinWatson 

R Square Change 
F Change 
df1 
df2 
Sig. F Change 
1.537 

1 
.879(a) 
.772 
.761 
.011 
.772 
66.189 
2 
39 
.000 

a. Predictors: (Constant), Hi, IAi ; b. Dependent Variable: TEAi  
Source: Own table, own calculations 
Table 4.17: ANOVA (SPSS) (b) – TEA (Romania)  
Model 
Sum of Squares 
df 
Mean Square 
F 
Sig. 

1  Regression 
.015 
2 
.008 
66.189 
.000(a) 
Residual 
.004 
39 
.000 

Total 
.020 
41 

a. Predictors: (Constant), Hi, IAi ; b. Dependent Variable: TEAi  
Source: Own table, own calculations 
Table 4.18: Coefficients (SPSS) (a) – TEA (Romania)  
Model 
Unstandardized 
Standardized 
t 
Sig. 
95 % Confidence 

B 
Std. Error 
Beta 
Lower B.  Upper B.  
1  (Constant) 
.050 
.003 
14.399 
0.000 
043 
.057  
IAi 
.004 
.001 
.447 
4.728 
0.000 
.002 
.006 

Hi 
2.00E007 
.000 
.538 
5.699 
0.000 
.000 
.000 

a. Dependent Variable: TEAi  
Source: own table, own calculations 
Having in mind that the impact of infrastructure did not seem throughout linear with regard to FDI and total economic activity it becomes clear that the linear models above underestimate the role of infrastructure. Thus, a curve estimation shall show how much the fit of a simple regression could be improved if a nonlinear model would be applied.
Table 4.19 : Curve Estimation for IAi and FDI (Romania)  
Model  Predictor 
R² 
Regression Fit F 
Constant 
B 
Beta 
Tvalue 
Sig. 
Multiple, linear, with Hi  IAi 
.821 
89.589 
.000 
.001 
.3 
3.583 
.001 
Simple, linear  IAi 
.504 
40.654 
–.003 
.002 
.710 
6.376 
.000 
Simple, cubic  IAi 
.852 
72.809 
–.001 
.001 
3 .358a 
3.890 
.000 
a. whole function: FDIi = –.001 + 1.187 IAi – 3.612 IAi + 3.358 IAi (with the positive terms significant)  
Source: own table, own calculations 
A similar estimation for total economic activity leads to the conclusion that the linear model is more appropriate as the individual significance of the equation parts gets lost in the cubic model.
Footnotes
[1] This might also apply to Dolj (DJ) in the South. The county is infrastructural relatively well endowed, but surrounded by less well endowed neighbors and thus seems like an isolated island on the map in Figure 4.4.
[2] Note: not all roads visible on this map are European Roads. Some of them are normal national roads.
[3] The regression result including Bucharest and Ilfov is even better, despite the considerable negative effect of Ilfov for the regression: y = 10517 + 0.0794x with R² = 0.584 (cf. Appendix 5 for the figure).