International Journal of Scientific & Engineering Research, Volume 4, Issue 7, July-2013 2269
ISSN 2229-5518
Partial Mantel Analysis on Estimating the
Resembance of Students Performance
Aronu, C. O, Ebuh, G. U., Ogbogbo G. O., Bilesanmi, A. O.
Nnamdi Azikiwe University, Awka - Nigeria
Abstract
The partial Mantel analysis is a test statistic that is used to measure the resemblance between two distance matrices after controlling for the effect of the third distance matrix measured over the same objects. The method used in this study is; permute the objects in one of the vectors (or matrices). Association on student’s performance of three faculties (Physical Sciences, Biosciences and Engineering) in Nnamdi Azikiwe University, Awka-Nigeria, was used to illustrate the method, where interest is on estimating the resemblance between two faculties students’ performance while controlling for the effect of the third faculty student performance. From the result obtained in this study, we conclude that there exists a strong negative resemblance between the performance of students in faculty of Physical sciences and faculty of Biosciences while controlling for the effect of student performance of faculty of Engineering for 100,000 permutations and using the Canonical distance which is “method three” in the “dist.quant (distance quantity)” function. R 2.13.0 programming package was used to run the analysis for 100,000 permutations.
Key words: Faculty, Canonical distance, Matrices, Permutations, Vectors, Engineering, Biosciences
—————————— ——————————
ultivariate tables of observations are usually condensed into resemblance matrices among any sampling unit of interest computed using similar-
Considering proximity matrices A, B , and C com- puted for three univariate or multivariate data tables.
ity distance (also called dissimilarity). Suppose we wish to consider three matrices, [1] proposed an ex-
The partial Mantel statistic, rM
(AB⋅C ), estimating
tension of the Mantel test to carry out partial correla- tion analysis in population genetics. [2], showed how French financial elite friendship ties are correlated
with (dis) similarity on several attribute variables,
such as political preference, educational institute, and club membership. In this study we measured the re-
the correlation between matrices A and B while con-
trolling for the effect of C , is computed in the same
way as a partial correlation coefficient:
semblance of student performance in three faculties and on three courses of an institution where interest is
rM (AB.C )=
rM (AB)− rM (AC )rM (BC )
(1)
on ascertain the performance association level of the selected students on three courses. Other contributors on Mantel and partial Mantel test includes; [3]; [4]; [5];
1− rM
(AC )2 1− r
(BC )2
[6]; [7]; [8]; [9] and [10]. The R- programming pack- age was used to run the analysis because it has the
1. Compute the Mantel correlations measure
ability of running mantel and partial mantel statistic for large number of permutations.
rM (AB),
rM (AC )
and rM (BC ). Cal-
2 Material and methodology
A partial Mantel test is a first-order partial correlation
analysis conducted on three distance matrices [1].
IJSER © 2013
culate the reference value of the of the test sta- tistic, rM (AB⋅C ), using Eq. 1.
2. Permute Aat random using matrix permutation algorithm to obtain A* .
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 4, Issue 7, July-2013 2270
ISSN 2229-5518
3. Compute r
(A* B)
and r
A*C , using
39, 50, 53, 65, 37, 41, 38, 41, 57, 71, 70, 65, 66, 71,
M M
83, 76, 62, 57, 81)
the value
rM (BC )
calculated in step 1,
R> MCB <-c(60, 84, 84, 85, 77, 69, 80, 80, 75, 87, 77,
78, 76, 62, 62, 59, 66, 67, 66, 59, 38, 53, 52, 34, 56,
compute rM
(A* B⋅C )using Eq.1 to obtain a
53, 33, 51, 34, 46)
R> BCH <-c(66, 66, 69, 60, 83, 89, 86, 71, 62, 80, 65,
value r*M of the partial correlation statistic un- der permutation.
4. Repeat step 2 and 3 a large number of times to
65, 75, 67, 66, 61, 73, 66, 62, 68, 32, 37, 47, 49, 49,
53, 34, 40, 36, 40)
R> ZOO <-c(56, 68, 89, 56, 86, 81, 57, 78, 89, 63, 64,
obtain the distribution of
* under permuta-
M
66, 67, 79, 68, 62, 60, 68, 78, 67, 51, 37, 31, 32, 43,
38, 55, 44, 42, 46)
tion. Add the reference value rM
to the distribution.
(A* B⋅C )
R> MECH <-c(47, 64, 59, 36, 31, 30, 56, 44, 24, 28,
58, 69, 80, 80, 72, 62, 55, 57, 77, 78, 64, 63, 85, 78,
83, 64, 83, 68, 68, 60 )
5. To determine the probability For a one – tailed test involving the upper tail, calculate the
probability as the proportion of values *
M
greater than or equal to rM . In the lower tail,
the probability is the proportion of
R> CIVIL <-c(21, 57, 63, 45, 22, 25, 43, 34, 35, 63,
71, 56, 86, 76, 76, 57, 54, 63, 57, 61, 75, 87, 63, 68,
89, 68, 61, 81, 72, 60)
R> ELECT <-c(40, 57, 64, 40, 32, 55, 67, 29, 46, 79,
81, 86, 80, 73, 56, 75, 74, 86, 84, 92, 63, 76, 64, 81,
81, 90, 86, 73, 65, 60)
R> FACULTYPHYSICALSCIENCES <-
ues *
M
smaller than or equal to rM .
matrix(c(STAT, MATHS, PHY), nrow = 3, byrow = TRUE)
R> FACULTYBIOSCIENCES <-matrix(c(MCB, BCH, ZOO), nrow = 3, byrow = TRUE)
The data for this study was presented as Appendix 1
3.0 Data analysis
Testing the hypothesis;
R> FACULTYENGINEERING <-matrix(c(MECH, CIVIL, ELECT), nrow = 3, byrow = TRUE)
It is important to note that the class distance of matrices FACULTYPHYSICALSCIENCES, FAC- ULTYBIOSCIENCES and FACULTYENGINEER-
H 1 +: Vs
H 2 +:
r (AB.C )= 0
r (AB.C )≠ 0
ING as defined above are based on canonical measure
(Method=1), labelled as FACULTYPHYSI- CALSCIENCESdist, FACULTYBIOSCIENCESdist
Inputting the data in Table 1on R 2.13.0 command
window, where STAT, MATHS and PHY are in FAC- ULTYPHYSICALSCIENCES matrix (matrix A), MCB, BCH and ZOO are in FACULTYBIOSCI- ENCES matrix (matrix B) while MECH, CIVIL and ELECT are in FACULTYENGINEERING matrix (matrix C) as given;
R>STAT <-c(78, 74, 68, 77, 78, 54, 75, 73, 56, 72, 61,
39, 55, 53, 50, 58, 48, 39, 64, 41, 79, 73, 67, 62, 71,
87, 70, 68, 69, 67)
R>MATHS <-c(53, 76, 69, 59, 78, 57, 76, 55, 57, 54,
66, 62, 39, 61, 38, 43, 65, 43, 55, 39, 72, 83, 77, 58,
57, 71, 80, 83, 81, 82 )
R> PHY <-c(66, 62, 69, 65, 78, 73, 70, 66, 66, 78, 67,
and FACULTYENGINEERINGdist respectively.
R> FACULTYPHYSICALSCIENCESdist <- dist.quant(FACULTYPHYSICALSCIENCES, method
= 1)
R> FACULTYBIOSCIENCESdist <- dist.quant(FACULTYBIOSCIENCES, method = 1)
R> FACULTYENGINEERINGdist <- dist.quant(FACULTYENGINEERING, method = 1)
Below is the elements of distance matrices FACUL- TYPHYSICALSCIENCESdist which contains objects
of matrix FACULTYPHYSICALSCIENCES on a class distances based on the canonical measure (meth-
IJSER © 2013 http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 4, Issue 7, July-2013 2271
ISSN 2229-5518
od =1). Where the result displayed by FACUL- TYPHYSICALSCIENCESdist expressed that the dis- tance between the performance of STAT and MATHS is 68.24222, STAT and PHY is 55.56078 and MATHS and PHY is 76.31514. R>FACULTYPHYSICALSCIENCESdist
STAT MATHS
MATHS 68.24222
PHY 55.56078 76.31514
Similarly, below is the elements of distance matrices FACULTYBIOSCIENCESdist which contains objects of matrix FACULTYBIOSCIENCES on a class dis- tances based on the canonical measure (method =1). Where the result displayed by FACULTYBIOSCI-
0)
Mantel statistic r: -1
Significance: 0.67327
Empirical upper confidence limits of r:
90% 95% 97.5% 99%
1 1 1 1
Based on 100 permutations, stratified within 0
4. Interpretation:
From the result obtained we observe that the partial
mantel measure of FACULTYPHYSICALSCIENC- ESdist, FACULTYBIOSCIENCESdist, while control- ling for the effect of FACULTYENGINEERINGdist
= -1 and a significance value = 0.67327 for 100,000
permutations. This expression can equally be ex-
ENCESdist expressed that the distance between the
performance of MCB and BCH is 56.92100, MCB and
pressed as given
r (AB.C )= −1
and 67.33% risk of
ZOO is 74.17547 and BCH and ZOO is 69.58448.
R> FACULTYBIOSCIENCESdist
MCB BCH
BCH 56.92100
ZOO 74.17547 69.58448
Similarly, below is the elements of distance matrices FACULTYENGINEERINGdist which contains ob- jects of matrix FACULTYENGINEERING on a class distances based on the canonical measure (method
=1). Where the result displayed by FACULTYENGI- NEERINGdist expressed that the distance between the performance of MECH and CIVIL is 75.51159, MECH and ELECT is 90.98351 and CIVIL and ELECT is 92.05433.
R> FACULTYENGINEERINGdist
MECH CIVIL CIVIL 75.51159
ELECT 90.98351 92.05433
The mantel.partial function was used to perform the
partial mantel test for 100,000 permutations, where “permutation” represents the number of permutations; R>mantel.partial(FACULTYPHYSICALSCIENCESdi
st, FACULTYBIOSCIENCESdist, FACULTYENGI-
NEERINGdist, method ="pearson", permutations =
100,000)
Partial Mantel statistic based on Pearson's product- moment correlation
Call:
mantel.partial(xdis = FACULTYPHYSI-
CALSCIENCESdist, ydis = FACULTYBIOSCIENC- ESdist, zdis = FACULTYENGINEERINGdist, method = "pearson", permutations = 100, strata =
rejecting the null hypothesis while it is true, which
fall’s on the acceptance region assuming α=0.05.
Where,
A=FACUTYPHYSICALSCIENCES, B=FACULTYBIOSCIENCESdist and C=FACULTYENGINEERINGdist.
5.0 Conclusion
From the interpretation we can conclude that there ex-
ists a strong negative resemblance between the per- formance of students in faculty of Physical science and faculty of Biosciences while controlling for the effect of performance of Faculty of Engineering for
100,000 permutations and using the canonical distance which is “method =1” in the “dist.quant” function. This implies that the class distance measures of the control which is Faculty of Engineering is far better than the measures of Faculty of Physical Sciences and Faculty of Biosciences as can be observed that in the Analysis section 3.0; hence the performance of the department in Faculty of Engineering is more associ- ated than that of other departments.
[1] Manly, B. J. F. Randomization and Regression Methods for Testing for Associations with Geograph- ical, Environmental and Biological Distances between Populations. Res. Popul. Ecol., 28, 201 – 218; 1986.
[2] Kadushin, C. Friendships among the French Fi- nancial Elite. American Sociological Review, (1995),
IJSER © 2013 http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 4, Issue 7, July-2013 2272
ISSN 2229-5518
60(2), 201–221.
[3] Aronu, C. O, Ebuh, G. U. Application of Mantel’s Permutation Technique on Asphalt Production in Ni- geria. International Journal of Statistics and Applica- tions 2013, 3(3): 81–85.
[4] Anderson, M. J. and Robinson, J. Permutation
Tests for Linear Models. Aust. N. Z. J.Stat, (2001)
43(1) 75-88.
[5] Baker, F. & Hubert, L. “The Analysis of Social Interaction Data” Sociological Methods and Research, (1981); 9, 339–361.
[6] Jackson, D. A. and Somers, K. M. “Are Probability Estimates from the Permutation Model of Mantel's Test Stable?” Canadian Journal of Zoology, (1989);
67(766-769).
[7] Krackhardt, D. Predicting with Networks: Nonpar- ametric Multiple Regression Analysis of Dyadic Data. Soc. Networks, (1988) 10, 359 – 381.
[8] Krackhardt, D. and Kilduff, M. “Friendship Pat- terns and Culture: The Control of Organizational Di- versity”. American Anthropologist, (1990), 92, 142–
154.
[9] Legendre, P. Qualitative methods and biogeo- graphic analysis; Evolutionary Biogeography of the Marine Algae of the North Atlantic (eds Garbary DJ & South RG), Vol. G22, pp. 9 – 34; 1990. NATO ASI Series, Springer – Verlag, Berlin.
[10] Manly, B. J. F. Randomization, Bootstrap and
Monte Carlo Methods in Biology (Second Edition). London: Chapman and Hall, 1997.
[11] Legendre, P. Comparison of Permutation Methods for the Partial Correlation and Partial Mantel Tests. J. Statist. Comput. Simulation, (67), 37 – 73; 2000.
IJSER © 2013 http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 4, Issue 7, July-2013 2273
ISSN 2229-5518
FACULTY PHYSICAL SCIENCES | FACULTY BIOSCI- ENCES | FACULTY ENGINEERING | |||||||
COURSE/DEPARTMENTS | STAT | MATHS | PHY | MCB | BCH | ZOO | MECH | CIVIL | ELECT |
GSS 101 | 78 | 53 | 66 | 60 | 66 | 56 | 47 | 21 | 40 |
74 | 76 | 62 | 84 | 66 | 68 | 64 | 57 | 57 | |
68 | 69 | 69 | 84 | 69 | 89 | 59 | 63 | 64 | |
77 | 59 | 65 | 85 | 60 | 56 | 36 | 45 | 40 | |
78 | 78 | 78 | 77 | 83 | 86 | 31 | 22 | 32 | |
54 | 57 | 73 | 69 | 89 | 81 | 30 | 25 | 55 | |
75 | 76 | 70 | 80 | 86 | 57 | 56 | 43 | 67 | |
73 | 55 | 66 | 80 | 71 | 78 | 44 | 34 | 29 | |
56 | 57 | 66 | 75 | 62 | 89 | 24 | 35 | 46 | |
72 | 54 | 78 | 87 | 80 | 63 | 28 | 63 | 79 | |
GSS 102 | 61 | 66 | 67 | 77 | 65 | 64 | 58 | 71 | 81 |
39 | 62 | 39 | 78 | 65 | 66 | 69 | 56 | 86 | |
55 | 39 | 50 | 76 | 75 | 67 | 80 | 86 | 80 | |
53 | 61 | 53 | 62 | 67 | 79 | 80 | 76 | 73 | |
50 | 38 | 65 | 62 | 66 | 68 | 72 | 76 | 56 | |
58 | 43 | 37 | 59 | 61 | 62 | 62 | 57 | 75 | |
48 | 65 | 41 | 66 | 73 | 60 | 55 | 54 | 74 | |
39 | 43 | 38 | 67 | 66 | 68 | 57 | 63 | 86 | |
64 | 55 | 41 | 66 | 62 | 78 | 77 | 57 | 84 | |
41 | 39 | 57 | 59 | 68 | 67 | 78 | 61 | 92 | |
MAT 102 | 79 | 72 | 71 | 38 | 32 | 51 | 64 | 75 | 63 |
73 | 83 | 70 | 53 | 37 | 37 | 63 | 87 | 76 | |
67 | 77 | 65 | 52 | 47 | 31 | 85 | 63 | 64 | |
62 | 58 | 66 | 34 | 49 | 32 | 78 | 68 | 81 | |
71 | 57 | 71 | 56 | 49 | 43 | 83 | 89 | 81 | |
87 | 71 | 83 | 53 | 53 | 38 | 64 | 68 | 90 | |
70 | 80 | 76 | 33 | 34 | 55 | 83 | 61 | 86 | |
68 | 83 | 62 | 51 | 40 | 44 | 68 | 81 | 73 | |
69 | 81 | 57 | 34 | 36 | 42 | 68 | 72 | 65 | |
67 | 82 | 81 | 46 | 40 | 46 | 60 | 60 | 60 |
Source: Nnamdi Azikiwe University, Awka Departmental student records for 2012 session
Key: STAT= Statistics department students, MATHS= Mathemat- ics department students, PHY= Physics department student, MCB= Microbiology department students, BCH=Biochemistry department students, ZOO= Zoology department students, MECH= Mechanical engineering department students, CIVIL= Civil engineering department student, ELECT= Electrical engi- neering department students, GSS =General social studies and MAT= Mathematics.
IJSER © 2013 http://www.ijser.org