SDA BOCCONI -­‐ GEMBA8 CLASS 2015

QUANTITATIVE METHODS

PERSONAL ASSIGNMENT

DATA ANALYSIS

BY

LORENZO CORONATI

Prof. Maurizio Poli

Via Bocconi 8

Office room: 517 (5th floor)

E-­‐Mail: maurizio.poli@sdabocconi.it

1

1. PRELIMINARY ANALYSIS

The main scope of the work and the data analysis consist in developing a multiple linear regression model capable of demonstrate the function between ITC cost and the selected independent variables. All data in this work have been extrapolated from Dataset Eurostat Datawherehouse.

The statistical units that have been studied are the 15 countries of the European Community as described in table 1. It has been utilized for the analysis a software called JMP provided by SDA Bocconi, University Bocconi Milan, with regular licence of use.

EU COUNTRIES Belgium Germany Greece Spain France Italy Hungary Netherlands Austria Poland Portugal Romania Finland Sweden United Kingdom

Table 1: European countries selected for the data analysis

The dependent variable studied consists in ITC expenditures measured in millions of Euros. The independent variables that have been chosen are the following:

ITC HIGH TECH PV CONSUMER VENTURE CAPITAL R&D

EMPLOYMENT

EMPLOYMENT IN ITC BROADBAND FIRMS BROADBAND HOUSE

Table 2 Dependent variables

The statistical summary of all variables has been summarized in figure n.1

2

Figure 1 Statistical Syntesis

1.
CORRELATION
MATRIX

Figure n. 2 explain the correlation matrix of all variables under our analysis. The ITC dependent variable shows us a very high direct correlation with the following variables: High Tech, Venture Capital, Employment, Employment ITC Broadband Firm; meanwhile it shows a very high indirect correlation with the Consumer and R&D variables

The correlation analysis among independent pairs of variables shows a high direct correlation between High Tech and Employment variables and also an extreme correlation between High Tech and Employment ITC. This correlation could create several multicollinearity problems of the multivariate model analyzed.

In the same time the variable Consumer appears to be negatively correlated with the Employment variable and also with the R&D variable: as it has been mentioned before even this case could create multicollinearity problems.

The outliers analysis has been summarized in figure n. 2 both with Mahalanobis distances and Jackknife distances and it doesn’t show specific outliers (or any particular value).

This means that it is possible to maintain the original data.

3

Figure 2 correlation matrix and outliers analysis

4

2.
BACKWARD
ANALYSIS ( MULTIPLE LINEAR REGRESSION).

The first multiple linear regression model consists by 1 (one) dependent variable (y) and 8 (eight) independent variables.

The estimated coefficient values obtained utilizing the software JMP can be summarized in figure n. 3.

Figure 3 Regression model with 8 variables independent

Considering the level of significance of 5%, designated as ALPHA (alpha=0,05), the significant statistic values have been associated to the independent variables High Tech, Consumer, Venture Capital and R&D. Instead the others are not coefficient significative. The p-­‐Value associated among Employment variable is 0,9673. This is the reason why the Employment variable has been excluded from the model. The resulting estimate of parameters can be summarized in figure n. 4.

5

Figure 4 Regression model with 7 variables independent model

Considering the level of significance of 5%, designated as ALPHA (alpha=0,05), the significant statistic values have been associated to the to the independent variables High Tech, Consumer, Venture Capital and R&D. Instead the others are not significative coefficients . The p-­‐Value associated among broadband House variable is 0,7378; for this reason the Broadband House variable has been excluded from the model again. The resulting estimate of the parameters can be summarized in figure n. 5.

6

Figure 5 Regression model with 6 variables independent

Considering the level of significance of 5%, designated as ALPHA (alpha=0,05), the significant statistical values have been associated to the to the independent variables High Tech, Consumer, Venture Capital, R&D and Employment ITC. Instead, it is not significative the coefficient associated to the Broadband Firms and its p-­‐Value associated is 0,1413; this is the reason why Broadband Firms variable has been excluded from the model. The resulting estimate of the parameters can be summarized in figure n. 6.

7

Figure 6 Regression model with 5 variables independent

Considering the level of significance of 5%, designated as ALPHA (alpha=0,05), the significant statistical values have been associated to the independent variables High Tech, Venture Capital and R&D. Broadband Firms and Consumer are not significant coefficients and Consumer has a P-­‐Value of 0,5157; this is the reason why Consumer variable has been

8

excluded from the model. The resulting estimate of the parameters can be summarized in figure n. 7.

Figure 7 Regression model with 4 variables independent

Considering the level of significance of 5%, designated as ALPHA (alpha=0,05), the significant statistical values have been associated to the independent variables High Tech, Venture Capital and R&D. It is not significant the coefficient associated to the Broadband Firms variable with a p-­‐Value of 0,6381. This is the reason why Broadband Firms variable has been excluded from the model. The resulting estimate of the parameters can be summarized in figure n. 8.

9

Figure 8 Regression model with 3 variables independent

CONCLUSION

Therefore, from the original eight independent variables took into consideration, only three explicate the variable dependent ITC. In particular, ITC variable is positively correlated with High Tech variable and also with the Venture Capital variable. This means that, when we have an increase of High Tech expenditures and Venture Capital investments, it is possible to estimate an increase in ITC investments. In contrast from our expectations, the regression model shows inverse correlation between ITC and R&D variable. This means that, when there is an increase of R&D expenditure, there is a decrease on ITC investments.

The linear regression model estimated is characterized by R-­‐Squared (our coefficient of determination or better the “power” – reliability -­‐ of the regression line model given by the coefficient above) of 0,9237. As a result the multivariate model explicate the 92,37 of IT variability. For this reason the predicted values in table 3 are characterized by low residuals.

COUNTRY ITC HIGH TECH PV V. Cap. Belgium 1400 6797000 0,134 Germany 2600 33001800 0,193 Greece 1200 2687400 0,007 Spain 2800 23903900 0,237 France 2600 29504400 0,308 Italy 2600 26617200 0,058 Hungary 1200 3879600 0,047 Netherlands 1800 7720100 0,226 Austria 1500 7416200 0,045 Poland 2700 17129500 0,142 Portugal 1400 4494500 0,118 Romania 1200 2622300 0,065 Finland 1300 3254400 0,247 Sweden 1900 8101300 0,896 United Kingdom 3800 25000900 1,128 Table 3 predicted and residual values

R&D INTRA

690,7 855,1 120,9 313,8 672,3 331,6 112,4 657,1 965,9 68,6 260,8 28,2 1302,7 1270,8 491,6

Res.l ITC -­‐37,0 -­‐242,7 -­‐113,0 164,8 -­‐265,0 32,5 -­‐226,7 193,3 249,4 433,5 -­‐78,1 -­‐215,7 184,2 -­‐236,3

Pred.ITC 1437,0 2842,7 1313,0 2635,2 2865,0 2567,5 1426,7 1606,7 1250,6 2266,5 1478,1 1415,7 1115,8 2136,3

156,8

3643,2

U. 95% L. 95% Mean Mean ITC ITC 1252,7 1621,3 2496,3 3189,1 1065,7 1560,3 2421,0 2849,4 2602,4 3127,6 2302,1 2832,9 1187,1 1666,2 1441,0 1772,5 985,7 1515,5 2043,3 2489,6 1270,9 1685,4 1148,4 1683,1 775,8 1455,9 1752,2 2520,3 3184,5

4101,8

10