 Open Access
 Total Downloads : 61
 Authors : Andriamahazoarivo Alex, Ravalison Francois
 Paper ID : IJERTV5IS120182
 Volume & Issue : Volume 05, Issue 12 (December 2016)
 DOI : http://dx.doi.org/10.17577/IJERTV5IS120182
 Published (First Online): 28122016
 ISSN (Online) : 22780181
 Publisher Name : IJERT
 License: This work is licensed under a Creative Commons Attribution 4.0 International License
Survey of the Causality between Military Training Quality and Security: the case of Malagasy Military
Andriamahazoarivo Alex Ministry of Defense Madagascar
Ravalison FranÃ§ois, Rakotomaria Etienne University of Antananarivo Madagascar
Abstract The impact of military training on security is not clear and deserves detailed empirical analysis in order to discern which one (if any) of the effects prevails This paper has focused on the relationship between military training quality and security. A Grangercausality analysis has been carried out in order to assess whether there is any potential predictability power of one indicator for the other. The conclusion that can be drawn is that military training quality can be used in order to predict security, but the opposite it is not true.
Key words: Granger causality Test, military training quality, security

INTRODUCTION
The relationship between military spending and economic growth has attracted the interest of many economists and researchers since Benoits study [1]. Since then, scholars have carried on several studies which, among others, are to investigate the presence and direction of the relationship between these two variables, either at the individual country level or through the examination of large samples of countries. The present study is about military training quality and security. The first variable is under military spending and the second variable is determinant for economic growth. So the research question is what is the relation between military training quality and security?.

METHODOLOGY
Data
Data used in this study are annual basis which cover the period of 20092015 and the variables are Malagasy military training quality (mtq) [3] and security (sec):

Data related to (mtq) were collected from the Centre de Perfectionnement des Officiers (or Improvement Center for Officers) and Ecole dEtatMajor (or Army Staff School). Quality of each training syllabus was evaluated on 15 Likert scale.

And we have carried out a survey among population to collect data corresponding to security. For that
Thus, the fundamental equation for the present study can be shown as follows:
() = () (1)
Where security (sec) is function of military training quality (mtq).
Granger causality tests
Granger (1969) proposed a timeseries data based approach in order to determine causality. In the Grangersense (mtq) is a cause of (sec) if it is useful in forecasting (sec). In this framework useful means that (mtq) is able to increase the accuracy of the prediction of (sec) with respect to a forecast, considering only past values of (sec).
There are three different types of situation in which a Grangercausality test [2] can be applied:

In a simple Grangercausality test there are two variables and their lags.

In a multivariate Grangercausality test more than two variables are included, because it is supposed that more than one variable can influence the results.

Finally Grangercausality can also be tested in a VAR framework; in this case the multivariate model is extended in order to test for the simultaneity of all included variables.
The empirical results presented in this paper are calculated within the first type that is a simple Grangercausality test in order to test whether military training quality Granger cause security and vice versa. This test is performed in order to identify the direction of the causal relationship between (sec) and (mtq). The casual relationships can be either unidirectional or bidirectional. This test estimates the following equations assuming there is no correlation between u1t and u2t
() = 1 () + 1 () +
= =
survey, we were based on indicators formulated in the Plan National de DÃ©veloppement (PND) (or National Plan Development). In that national document, security is framed by three indicators: homicidal ratio, ratio of military frequency per population, and percentage of people feeling in security.
1 (2)
=
=
() = =1 () + 1 () + 2
(3)
Based on the coefficients for the equations (2) and (3) four different hypotheses about relationship between (sec) and (mtq) can be formulated:

Unidirectional Grangercausality from (mtq) to (sec). In this case (mtq) increases the prediction of the (sec) but
not vice versa. Thus 1 0 and 1 = 0

(sec) is regressed on its past values excluding (mtq) in the regressors. This is called the restricted regression, from which we obtain the restricted sum of squared residuals.

Thus, a second regression is computed including the lagged SP. This is called the unrestricted regression from
= =


Unidirectional Grangercausality from (sec) to (mtq). In this case the (sec) increases the prediction of (mtq) but
not vice versa.. Thus 1 = 0 and 1 0
which the unrestricted sum of squared residuals is obtained.
=
=
The statistics is defined as:

Bidirectional (or feedback) causality. In this case
=
=
=
=
1 0 and 1 0, so in this case the (sec)
increases the prediction of (mtq) and vice versa

Independence between (sec) and (mtq). In this case there
=
=
=
is no Granger causality in any direction, thus 1 =
( + + 1)
=
=
0 and 1 = 0
Hence by obtaining one of these results it seems possible to detect the causality relationship between military training quality and the security in Madagascar as far as assessment of strategic management is concerned [4].


RESULTS AND DISCUSSION
Figure 1: Scatter plots for (sec) and (mtq)
Figure 1 shows how just a quick view on the data can support a positive relation between the two variables (in percentages). The analysis in this paper will show in formal terms what kind of relation can be hypothesized on these two variables.
The first step in this analysis concerns the stationarity of the (sec) and (mtq) series. Granger causality requires that the series have to be covariance stationary, so an Augmented DickeyFuller or ADF test has been calculated. For all of the series the null hypothesis H0 of non stationarity can be rejected at a 5% confidence level.
Then, since the Grangercausality test is very sensitive to the number of lags included in the regression, both the Akaike (AIC) and Schwarz Information Criteria have been used in order to find an appropriate number of lags.
After that these requirements have been satisfied, Granger causality tests are computed. Taking equation (2) as an example, the two steps procedure in testing whether (mtq) causes (sec) is as follows.
where SSRr and SSRu are the two sums of squared residuals related to the restricted and unrestricted form of the equation; the elements that form the degrees of freedom are T , that is the number of observations while n and m are the number of lags as it can be seen from (2). The same procedure is used in order to test for the inverse Grangercausality relation in (3).
It is important that the data are covariance stationary in order to perform any kind of such regression, given the key of interpretation that we are looking for. For this the ADF test has been performed. This is aclassic choice in literature and very strong test against unit roots. It is worth emphasizing that the two series that we are working with are already growth patterns, therefore we expect them to be I(0).The result reflects the I(0) state of the variables. It is also possible to see this result from the graphs above that show the rates of growth of the two series.
Since the series are covariance stationary we can proceed to checking f or the number of lags to input in the model. The Granger causality test is sensitive to this kind of formatting of the model, and it is therefore important to choose and information criterion to base the decision on the number of lags to apply to the two series in the regressions to follow. For this purpose we have analyzed a large range of lags both for the referring to the (sec). Many previous works use the criterions of Akaike and Schwarz to formulate these choices. The optimal values are m = 2 and n = 7 form defined as the lag of the (sec) series and n the lag applied for the mtq series.
Thus, the results of Granger Causality for equations (2) and

are represented in table 1 and 2. The tables report the results corresponding to di erent regressions, in order to have a comparison of the di erent regressions outputs. The values of F statistic suggest that (mtq) Grangercauses (sec)2, and (sec) does not cause (mtq). Thus, it can be argued that past values of (mtq) contribute to the prediction of the present value of (sec) even with past values of (sec). Moreover by the single regressions it can be showed that also with 5 lags much of the coecients have positive sign and with an acceptable significance level. However it has to be taken in account that the level of R2 is low, reminding that past rates of (mtq) could have a limited ability for the prediction of (sec).
For the equation (2) the associated F tests give the opposite result, in fact there seems to be no GrangerCausality from past values of (sec) for future values of (mtq). It has to be noted that this holds for all the specifications tried, and so in this case the null hypothesis of no causality from (sec) to (mtq) . Moreover all the R is close to zero, and the Fratios (that test for all the righthand coecients significance) are statistically insignificant.
Concluding our tests for Granger causality reflects what showed and asassessed in the theory. There seems not to be any causality from (sec) to (mtq). But an inverse Granger causality seems to be possible even if the relationship does not seem to be so strong. Indeed this can be found in the current and past events, which showed more than once how the (mtq) is not always in tune with the (sec). However, to the extent that the variation in the (mtq) can be seen as a leading indicator for the fluctuations of the aggregate output of security, there is a better chance for countercyclical policies to be adopted in advance.
TABLE I: RESULTS OF GRANGERCAUSALITY TESTS EQUATION 2
m
n
DWStat
FStat
R2
FRatio
2
1
2.1182
5.573(1,120)
0.119
6.215(3,120)
2
2
2.0372
11.015(2,120)
0.210
9.127(4,120)
2
3
1.9583
7.983(3,120)
0.223
7.733(5,120)
2
4
1.8840
5.555(4,120)
0.216
6.119(6,120)
2
5
1.9582
3.885(5,120)
0.224
5.398(7,120)
2
6
2.0071
3.272(6,120)
0.230
4.801(8,120)
2
7
2.0042
3.043(7,120)
0.240
4.458(9,120)
TABLE II: RESULTS OF GRANGERCAUSALITY TESTS EQUATION 3

CONCLUSION
There is a causality relation between military training quality and security in Madagascar base of military learning and quality [5]. The Granger test has permitted to undertake the survey. The Malagasy case reveals that is that military training quality can be used in order to predict security, but the opposite it is not true.

REFERENCES


E. Benoit, Defense and economic growth in developing countries: Lexington Books: Lexington, MA, 1973

C. J. Granger C. J., Investigating Causal Relationships by Econometrics Models and Cross Spectral Methods, Econometrica, Vol. 37, 1969, pp. 425435.

J. L. Montes, J. L. et al., Factors affecting the relationship between total quality management and organisational performance, International Journal of Quality and Reliability Management, 2003, 20(2)

D. Jennings, Strategic management: an evaluation of the use of three learning methods, The Journal of Management Development, 21(9), pp. 655 665, 2002
 [5] R. McAdam et al., The links between organisational learning and total quality: a critical review, Journal of European Industrial Training, 22(2), pp. 47 56, 1998
m 
n 
DWStat 
FStat 
R2 
FRatio 
2 
1 
1.9122 
0.007(1,120) 
0.003 
0.161(3,120) 
2 
2 
1.9281 
0.647(2,120) 
0.012 
0.444(4,120) 
2 
3 
2.0166 
0.461(3,120) 
0.017 
0.471(5,120) 
2 
4 
2.0020 
0.307(4,120) 
0.014 
0.319(6,120) 
2 
5 
1.9841 
0.462(5,120) 
0.022 
0.425(7,120) 
2 
6 
2.0011 
0.484(6,120) 
0.270 
0.448(8,120) 
2 
7 
2.0028 
0.408(7,120) 
0.268 
0.389(9,120) 