Journal of Reliability and Statistical Studies; ISSN (Print): 0974-8024, (Online):2229-5666

Vol. 4, Issue 2 (2011): 105-117

APPLYING LOGISTIC REGRESSION MODEL TO THE
EXAMINATION RESULTS DATA
Goutam Saha
Govt. Degree College, Kamalpur, Tripura
E Mail: [email protected]
Abstract
The binary logistic regression model is used to analyze the school examination results
(scores) of 1002 students. The analysis is performed on the basis of the independent variables viz.
gender, medium of instruction, type of schools, category of schools, board of examinations and
location of schools, where scores or marks are assumed to be dependent variables. The odds ratio
analysis compares the scores obtained in two examinations viz. matriculation and higher
secondary.
Key Words: Logistic Regression Model, Correspondence Analysis, Odds Ratio, Likelihood
Ratio Tests, Wald Statistic, Shortest Confidence Interval.
1. Introduction
Neither two students nor two schools are identical. Students differ in gender,
culture, religion, language, home environment, financial status of parents etc., whereas
the schools differ in size of students, quality of teacher, infrastructure, location of the
school, aid provided by the government etc.. Obviously performance of the students
measured in terms of scores or grades obtained by them in examinations varies from
student to student and school to school. The variability in scores is a function of social
climate which has to be studied and analyzed scientifically. The history of analyzing the
students performance is as old as history of education. However formal presentation of
analysis started around early thirties of the 20
th
century.
The performance measure corresponding to different independent variables
may be analyzed using logistic regression analysis. Logistic regression has been
successfully employed in social science, biostatistics, genetics and demographic issues,
but as far as school examinations are concerned, not many research articles are
available.
This paper deals with presentation and analysis of examination results of
Tripura: north-east India. The matriculation results of the students admitted in higher
secondary (science stream) are collected, along with their higher secondary scores. The
data collected consist of grades of the students in both the examinations along with their
scores in English, Mathematics and Science subjects.
It is assumed that scores of students are affected by social environment
controlled by the parameters viz. (i) gender (male, female), (ii) medium of instruction
(English, Bengali), (iii) type of schools (boys, girls, and co-educational), (iv) category
of schools (Govt., non-Govt.), (v) board of examinations i.e., Tripura Board of
Secondary Education (TBSE), Central Board of Secondary Education (CBSE) and
Indian Certificate of School Examination (ICSE) and (vi) location of schools (urban,
Journal of Reliability and Statistical Studies, December 2011, Vol. 4 (2) 106
rural). The logistic regression approach has been adopted to study the examination
scores under the variables mentioned above.
Scores of students are partitioned into two sets viz. [0, 45) and [45, 100].
Since in the above mentioned examination 45% and above (obviously 60% and above
means first class) marks indicate second class, hence the students are classified as
belonging to two different categories as far as their scores are concerned. As a
consequence idea of binary logistic regression analysis seems to be appropriate when
scores are functions of independent variables mentioned above.
A brief review of literature on this subject is included in section 2. Section 3
has data source. Materials and methods are shown on section 4. Section 5 contains
numerical analysis of binary logistic regression. Results and interpretations are
incorporated in section 6.
2. Review of Literature
Peng et al (2002) applied the logistic regression technique to compare the
sample data of gender and recommendation for remedial reading instruction. David et.
al. (2001) used logistic regression analysis to determine whether grade point average
and hours of education is significant predictor of performance on the national athletic
trainers association board of certificate examination. E. L. Dey and Astin A. W. (1993)
studied the focus on the practical implications of applying logistic regression, probit
analysis and linear regression to the problem of predicting the college student retention.
Jason et. al. (2001) analyzed the logistic regression method to predict the probability of
passing a course based on the scores on California chemistry diagnostic test at two
different institutions with two different instructors over multiple years. Robert B. and
Vaughan B. (2006) checked which factors were key in enabling or constraining a
students ability to close the achievement gap during the school results. Erin et al.
(2010) used multilevel logistic regression analyses, to explore the school and student
level characteristics associated with moderate and high levels of physical activity
among school students. In series of articles Sarma and Sarmah (1999), Saha and
Sarmah (2010) and Saha and Sarmah (2011) discussed the probabilistic analysis and
testing of some important hypothesis using Markov chain.
3. Data Source
1002 samples have been collected from the entire state of Tripura a part of
north-east India, through a Minor Research Project (MRP) entitled Prospects and
Problems of Educational development (Higher Secondary Stage) in Tripura - An in-
depth Study sponsored by University Grants Commission (UGC), New Delhi, India.
In this survey from a total population of 225 schools, offering both
matriculation and higher secondary (science stream) courses, a sample of 75 schools are
randomly selected and data related to examination scores are collected for analysis.
The results of the students in both the examinations are assumed to be influenced by the
variables gender, medium of instructions, type of schools, category of schools, board of
examinations and location of schools.
Applying Logistic Regression Model 107
4. Materials and Methods
The Model
Let us consider the indicator variable 1 Y
i
= , if
Y
i
e [45, 100]
= 0, if
Y
i
e [0, 45)
Now let us assume P
r
(Y
i
=1) = = 1 - P
r
(Y
i
=0)
Where, can be written as,
Where as,

= the constant of the equation and

= the coefficient of the

predictor variables X
i
for i = 1, 2, , n.
An alternative form of the logistic regression equation is:
(1)
The goal of logistic regression is to correctly predict the category of outcome
for individual cases using the most parsimonious model. To accomplish this goal, a
model is created that includes all predictor variables that are useful in predicting the
response variable.
Logistic regression calculates the probability of success over the probability of
failure; the results of the analysis are in the form of an odds ratio. The odds ratio is a
measure of effect size (Westergren A. 2001), describing the strength of association or
non-independence between two binary data values. It is used as a descriptive statistics,
and plays an important role in logistic regression. Unlike other measures of
association for paired binary data such as the relative scores, the odds ratio treats the
two variable being compared symmetrically, and can be estimated using some type of
non-random samples.
Hypothesis testing in logistic regression involves reasoning by contradiction.
The first assumption or the null hypothesis is that, the predictor coefficient is zero in the
population. Hypothesis test tell whether there is sufficient evidence in the sample data
to reject the null hypothesis and therefore to accept the alternative hypothesis that the
predictor variable coefficient differ from zero. Confidence intervals can be used for
hypothesis testing as well as for regression coefficients.
The odds ratio may be presented as
Journal of Reliability and Statistical Studies, December 2011, Vol. 4 (2) 108
l n
1
P
i
L
i
P
i
|
= |
- |

(2)
With [ 1] 1 [ 0] P P X P X
i r i r i
= = = - = , where X
i
is the independent variable
corresponding to i
th
category for i = 1, 2, , n.
5. Numerical Analysis of Binary Logistic Regression
In this paper binary logistic regression analysis is performed with dependent
variable of total marks obtained in higher secondary and matriculation examination, in
presence of independent variables gender, medium of instruction, type of schools, board
of examinations, category of schools and location of schools.
By applying the method of Correspondence Analysis (CA), it is observed that,
there exists a significant correlation between (i) type of schools and board of
examinations with r = 0.228, (ii) type of schools and category of schools with r = 0.141
and (iii) category of schools and board of examinations with r = 0.266. Here, we
perform binary logistic regression analysis.
6. Results and Interpretation
Data Variable Data Explanation Data Type Conditioned Used
Dependent Variable
Marks in higher
secondary /
matriculation
examination
Not-Satisfied
Satisfied
Binary 0 - Not Satisfied
1 - Satisfied
Independent Variables
Gender Gender of Student Binary 0 - Female
1 Male
Medium Medium of
Instruction
Binary 0 English
1 - Bengali
School Type Type of Schools Categorical
0 Boys
1 Girls
2 Co-Educational
Board Board of
Examinations
Categorical
0 TBSE
1 CBSE
2 - ICSE
Schools Category Category of Schools Binary 0 Govt.
1 Non-Govt.
Location Location of School Binary 0 Urban
1 - Rural
Table 1: Coding of Variables affecting Higher Secondary and Matriculation Results
Applying Logistic Regression Model 109
Variable Category Frequency %
Gender Female
(Male)
580
422
42.10
57.90
Medium English
(Bengali)
321
681
32.00
68.00
School Type Girls
Boys
(Co-Education)
204
214
584
20.40
21.40
58.30
Board TBSE
CBSE
(ICSE)
707
151
144
70.60
15.10
14.40
School Category Govt.
(Non-Govt.)
432
570
43.10
56.90
Location Urban
(Rural)
476
526
47.50
52.50
Table 2: Categorical Variables Affecting Higher Secondary and Matriculation Results
Here the reference category is shown in parenthesis (2
nd
column in Table 2)
whose odds ratio is 1.
Classification Frequency Percentage (%)
Not-Satisfied
Satisfied
135
867
13.50
86.50
Total 1002 100.00
Table 3: Classification of Total Marks in Higher Secondary Examination
Classification Frequency Percentage (%)
Not-Satisfied
Satisfied
053
949
05.30
94.70
Total 1002 100.00
Table 4: Classification of Total Marks in Matriculation Examination
By dropping one of the variables which are correlated, i.e., type of schools,
board of examinations and category of schools are not used simultaneously as
independent variables. Hence, we perform binary logistic regression analysis in three
stages. In the second column of the following tables B represents the coefficient for
the constant (also called the intercept) in the null model.
Journal of Reliability and Statistical Studies, December 2011, Vol. 4 (2) 110
6.1 Analysis (Stage I)
Here we consider higher secondary result as a dependent variable; where as
gender, medium of instruction, school type and location of schools are independent
variables.
95% C.I.for EXP(B)
B S.E. Wald df Sig. Exp(B)
Lower Upper
Gender (Female) 1.008 0.303 11.098 1 0.001 2.740 1.514 4.959
Medium (English) 0.808 0.243 11.089 1 0.001 2.244 1.395 3.611
School_Type 8.421 2 0.015
School_Type (Girls) -0.402 0.348 1.333 1 0.248 0.669 0.338 1.324
School_Type (Boys) 0.683 0.258 7.015 1 0.008 1.979 1.194 3.281
Location (Urban) 0.343 0.195 3.102 1 0.078 1.409 0.962 2.064
Constant 1.079 0.157 47.089 1 0.000 2.942
Table 5: Logistic Regression Analysis of Higher Secondary Examination: Overall Results
Discussion
The Wald statistic and the corresponding significance level test, the
significance of each of the covariate and dummy independent variables in the model are
shown in the above table. If the Wald statistic is significant (i.e., less than 0.05) then the
parameter is significant in the model. Of the independent variables, location of school is
insignificant, whereas gender of students, medium of instructions and type of schools
have significantly affected the results of students in higher secondary examination.
The performances of female students are approximately 3 times higher than
that of the performance of male students. As far as medium of instruction is concerned,
it can be seen that, the performance of English medium schools are 2.244 times better
than that of the Bengali medium schools.
As shown in the above table, type of schools (i.e., girls and boys schools) as
a whole is a significant factor with p value 0.015 corresponding to examination scores.
Also it is observed that, performance of girls and boys schools are 0.669 and 1.979
times better than the co-educational schools respectively. Similarly, the performances
of urban students are 1.409 times better than the rural students.
Looking at the length of confidence interval of estimated odds, we find that
school type (1) i.e., Girls schools are estimated with 95% confidence having shortest
interval length.
Applying Logistic Regression Model 111
Now we consider Matriculation result as a dependent variable, where as
gender of students, medium of instruction, school type and location of schools are
independent variables.
95% C.I.for EXP(B)
B S.E. Wald df Sig. Exp(B)
Lower Upper
Gender (Female) 1.150 0.503 5.222 1 0.022 3.157 1.178 8.463
Medium (English) -0.125 0.312 0.159 1 0.690 0.883 0.479 1.628
School Type 1.693 2 0.429
School_Type (Girls) -0.744 0.575 1.675 1 0.196 0.475 0.154 1.466
School_Type (Boys) -0.056 0.346 0.026 1 0.872 0.946 0.480 1.863
Location (Urban) -0.313 0.287 1.187 1 0.276 0.731 0.416 1.284
Constant 2.846 0.273 108.37 1 0.000 17.221
Table 6: Logistic Regression Analysis of Matriculation Examination: Overall Results
Discussion
Looking at the above Table 6, it is observed that, gender of students
contributes significantly to the model. Where as medium of instruction along with type
and location of schools are found to be insignificant in explaining the satisfactory
results in matriculation.
Now in particular, the performances of female students are 3.157 times higher
than that of the performance of male students. On the other hand for medium of
instruction, it can be seen that, the performance of English medium schools are 0.883
times better than that of the Bengali medium schools.
Also from the above table, type of schools (i.e., girls and boys schools) as a
whole is insignificant factor with p value 0.429 corresponding to examination scores.
Hence, it is observed that, performance of girls and boys schools are 0.475 and 0.946
times better than the co-educational schools respectively. Similarly, the performances
of urban students are 0.731 times better than the rural students.
Looking at the length of confidence interval of estimated odds, we find that
location of school is estimated with 95% confidence having shortest interval length.
Comparison of odds ratios of total marks obtained in higher secondary and
matriculation examination for the respective set of independent variables:
Journal of Reliability and Statistical Studies, December 2011, Vol. 4 (2) 112
Figure 1: Comparison of Odds Ratios obtained in Higher Secondary (HSSLC) and
Matriculation (HSLC) Examination
Discussion
It is apparent from the Figure 1, though female students are better performer
in matriculation, the performance of girls schools is worse in same examination
compared to their performance in higher secondary examination. Similarly English
medium and urban students show the better performance in higher secondary level
compared to matriculation. Boys schools also show better performance in higher
secondary stage.
6.2 Analysis (Stage II)
Here we consider Higher Secondary result as a dependent variable, where as
gender of students, medium of instruction, board of examination and location of schools
are independent variables.
95% C.I.for EXP(B)
B S.E. Wald df Sig. Exp(B)
Lower Upper
Gender (Female) 0.609 0.205 8.839 1 0.003 1.839 1.231 2.747
Medium (English ) -0.635 0.395 2.590 1 0.108 0.530 0.244 1.148
Board 16.696 2 0.000
Board (TBSE) -1.351 0.477 8.005 1 0.005 0.259 0.102 0.660
Board (CBSE) 0.994 0.556 3.203 1 0.073 2.703 0.910 8.032
Location (Urban)
0.450 0.195 5.347 1 0.021 1.568 1.071 2.297
Constant 2.582 0.490 27.751 1 0.000 13.225
Table 7: Logistic Regression Analysis of Higher Secondary Examination: Overall Results
0
0 . 5
1
1 . 5
2
2 . 5
3
3 . 5
1 2 3 4 5
I n d e p e n d e n t V a r i a b l e s
O
d
d
s

R
a
t
i
o
s
O d d s R a t i o s i n H S S L C
E x a m i n a t i o n
O d d s R a t i o s i n H S L C
E x a m i n a t i o n
Applying Logistic Regression Model 113
Of the independent variables, gender of students and location of schools
contribute significantly to the model. Board of examination turned out to be highly
significant along with TBSE and CBSE whereas CBSE is marginally significant to the
model. Only medium of instruction (English medium) shows insignificant result.
The performances of female students are 1.839 times higher than that of the
performance of male students. Whereas the performance of English medium schools are
0.530 times better than that of the Bengali medium schools.
As far as board of examination is concerned, it is observed that, TBSE and
CBSE boards are 0.259 and 2.703 times better than that of ICSE board. On the other
hand, urban students are 1.568 times better than that of rural students.
Looking at the length of confidence interval of estimated odds, we find that,
board (1) i.e., Tripura Board of Secondary Education (TBSE) is estimated with 95%
confidence having shortest interval length.
Now we consider Matriculation result as a dependent variable, where as
gender of students, medium of instruction, board of examination and location of schools
are independent variables.
95% C.I.for EXP(B)
B S.E. Wald df Sig. Exp(B)
Lower Upper
Gender (Female) 0.755 0.321 5.526 1 0.019 2.128 1.134 3.993
Medium (English) -0.618 0.579 1.139 1 0.286 0.539 0.173 1.678
Board 1.290 2 0.525
Board (TBSE) -0.733 0.672 1.192 1 0.275 0.480 0.129 1.792
Board (CBSE) -0.103 0.558 0.034 1 0.853 0.902 0.302 2.691
Location (Urban)
-0.288 0.286 1.011 1 0.315 0.750 0.428 1.314
Constant 3.513 0.701 25.120 1 0.000 33.555
Table 8: Logistic Regression Analysis of Matriculation Examination: Overall Results
From the above Table 8, it has been observed that, only gender contributes
significantly to the model. The remaining variables are found to be insignificant.
The performances of female students are 2.128 times higher than that of the
performance of male students. On the other hand, performance of English medium
schools is 0.539 times better than that of the Bengali medium schools.
As far as board of examination is concerned, it is observed that, TBSE and
CBSE boards are 0.480 and 0.902 times better than that of ICSE board. Similarly,
Journal of Reliability and Statistical Studies, December 2011, Vol. 4 (2) 114
performances of urban students are better than the rural students with 0.750 times.
Looking at the length of confidence interval of estimated odds, we find that,
location of school is estimated with 95% confidence having shortest interval length.
Comparison of Odds Ratio obtained in higher secondary and matriculation examination
for the respective set of independent variables.
Figure 2: Comparison of Odds Ratios obtained in Higher Secondary (HSSLC) and
Matriculation (HSLC) Examination
Discussion
Observing the Figure 2, we may conclude that, female performance is better in
matriculation examination than that of higher secondary examination. It is interesting to
know that, performance of English medium schools in both the examinations remain
same. Similarly performance of TBSE shows that better results in matriculation
comparing to higher secondary examination. Where as, CBSE schools are better
performer in higher secondary than that of matriculation examination. In fact urban
students are also showing better result in higher secondary examination.
6.3 Analysis (Stage III)
Here we consider Higher Secondary result as a dependent variable, where as
gender of students, medium of instruction, category of school and location of schools
are independent variables.
95% C.I.for EXP(B)
B S.E. Wald df Sig. Exp(B)
Lower Upper
Gender (Female) 0.564 0.203 7.732 1 0.005 1.758 1.181 2.617
Medium (English ) 0.730 0.241 9.192 1 0.002 2.075 1.294 3.326
Category of Schools (Govt) -0.349 0.190 3.375 1 0.066 0.705 0.486 1.024
Location (Urban)
0.391 0.193 4.099 1 0.043 1.479 1.013 2.159
Constant 1.464 0.178 67.934 1 0.000 4.323
Table 9: Logistic Regression Analysis of Higher Secondary Examination: Overall Results
0
0.5
1
1.5
2
2.5
3
1 2 3 4 5
I ndependent Var i abl es
Odds Rat ios in HSSLC Examinat ion
Odds Rat ios in HSLC Examinat ion
Applying Logistic Regression Model 115
From the above Table 9, it is observed that, all the variables except category of
school are found to be significant.
The performances of female students are 1.758 times higher than that of the
performances of male students. As far as medium of instruction is concerned, 2.075
times better performance of English medium school is observed, than that of the
Bengali medium schools.
On the other hand, the performances of Govt. schools comparing to the others
i.e., non-Govt. schools are better (with odds ratios 0.705). Similarly, the performances
of urban students are 1.479 times better than the rural students.
Looking at the length of confidence interval of estimated odds, we find that
category of school is estimated with 95% confidence having shortest interval length.
Here we consider Matriculation result as a dependent variable, where as gender of
students, medium of instruction, category of school and location of schools are
independent variables.
Table 10: Logistic Regression Analysis of Matriculation Examination: Overall Results
In third stage analysis, gender is also the only variable found to be
significantly contributing to the model.
The performances of female students are 2.113 times higher than that of the
performances of male students. Whereas performance of English medium school is
0.890 times better than that of the Bengali medium schools.
On the other hand, it is observed that, the performances of Govt. schools are
0.645 times better than, the non-Govt. schools. Similarly, the performances of urban
students are 0.719 times better than the rural students.
Looking at the length of confidence interval of estimated odds, here we also
find that category of school is estimated with 95% confidence having shortest interval
length.
95% C.I.for EXP(B)
B S.E. Wald df Sig. Exp(B)
Lower Upper
Gender ( Female) 0.748 0.320 5.467 1 0.019 2.113 1.129 3.955
Medium ( English ) -0.117 0.310 0.143 1 0.706 0.890 0.485 1.632
Category of School (Govt) -0.438 0.288 2.322 1 0.128 0.645 0.367 1.134
Location ( Urban)
-0.329 0.287 1.319 1 0.251 0.719 0.410 1.262
Constant 3.042 0.295 106.53 1 0.000 20.950
Journal of Reliability and Statistical Studies, December 2011, Vol. 4 (2) 116
Comparison of Odds Ratios obtained in higher secondary and matriculation
examination for the respective set of independent variables.
Figure 3: Comparison of Odds Ratios obtained in Higher Secondary (HSSLC) and
Matriculation (HSLC) Examination
Discussion
Observing the Figure 3, it may be point out that, females show better
performance in matriculation compared to higher secondary examination. But
performance of English medium schools turned out to be better in higher secondary
examination. On the other hand, Govt. schools are showing almost same performance
in both the examination. Urban schools shows 2 times better performance in higher
secondary examination compared to matriculation.
Overall Conclusion
From the above observations in all the three stages, we may conclude that,
females are always showing best performances in both the examinations. But their
performance is found to be still better in matriculation as shown in our analysis in all
three stages. The performances of English medium schools are found to be satisfactory
in higher secondary examination. Similarly urban schools always show better
performance in higher secondary examination.
Acknowledgement
The author acknowledges the University Grants Commission (UGC), New
Delhi, India for providing financial support through a Minor Research Project (MRP).
The author also expresses his gratitude to the editor and the reviewers of Journal of
Reliability and Statistical Studies (JRSS) to improve the paper through constructive
suggestions.
References
1. Agresti, A. (1996). An Introduction to Categorical Data Analysis, John Wiley
and Sons, Inc.
2. Hosmer, D. and Stanley, L. (1989). Applied Logistic Regression, John Wiley
and Sons, Inc.
3. Menard, S. (1995). Applied logistic regression analysis (Sage University Paper
Series on Quantitative Applications in the Social Sciences, 07106), Thousand
Oaks, CA: Sage.
0
0.5
1
1.5
2
2.5
1 2 3 4
I ndependent Var i abl es
Odds Rati os i n HSSLC Exami nati on
Odds Rati os i n HSLC Exami nati on
Applying Logistic Regression Model 117
4. Menard, S. (1995). Applied Logistic Regression Analysis. Sage
Publications.Series: Quantitative Applications in the Social Sciences, No. 106.
5. Menard, S. (2000). Coefficients of determination for multiple logistic
regression analysis, The American Statistician, 54(1), p. 1724.
6. Saha, G. and Sarmah, P. (2010). Statistical Analysis of School Examination
Result with Special Reference to the State of Tripura: North-East India,
Journal of Statistics Sciences, 2, p. 111-121.
7. Saha, G. and Sarmah, P. (2012): Stochastic Modeling of the Grading Pattern in
Presence of the Environmental Parameter, Electronic Journal of Applied
Statistical Analysis (Accepted).
8. Sarma, R. and Sarmah, P. (1999). A Stochastic Modeling on Grading System,
In Proceedings of the Second International Conference on Operations and
Quantitative Management in the Global Business Environment (ICOQM),
Ahmedabad, India, 36
th
January 1999, p. 276281.
9. Sarma R. and Sarmah P. (1999). Analysis of Results Based on Grades, In
Proceedings of the Second International Conference on Operations and
Quantitative Management in the Global Business Environment (ICOQM),
Ahmedabad, India, 36
th
January 1999, p. 282290.
10. Tabachnick, B. and Linda, F. (1996). Using Multivariate Statistics, Third
edition. Harper Collins.