![Ties Adjusted Nonparametric Statististical...
www.ijmsi.org 11 | Page
TABLE 3: Ranks assigned to students GPAs of Table 1 for use with the Friedman’s Test.
S/No Year 1 Year 2 Year 3 Year 4
1 3 1 2 4
2 2 1 3 4
3 2 1 4 3
4 4 2 1 3
5 2 1 3.5 3.5
6 3 1 4 2
7 2 1 3 4
8 3 2 1 4
9 4 1 3 2
10 3 2 1 4
11 2 1 3.5 3.5
12 2 1 4 3
13 2 1 3 4
14 2 1 4 3
15 2 1 3 4
16 2 1 3 4
17 1 2 3 4
𝑅.𝑗 41 21 49 59
Now using the sum of ranks given at the bottom of table 3 with the Friedman’s Two-Way Analysis of Variance
by Ranks test statistic given as
𝜒2
=
12
𝑛𝑐 𝑐 + 1
𝑅.𝑗
2
4
𝑗 =1
− 3𝑛 𝑐 + 1 − − − − − − − −(23)
we have
𝜒2
=
12 412
+ 212
+ 492
+ 592
17 4 5
− 3 17 5
=
12 1681 + 441 + 2401 + 3481
340
− 255 = 280.14 − 255
= 25.14 (𝑝 − 𝑣𝑎𝑙𝑢𝑒 = 0.000)
which with 3 degrees of freedom is also statistically significant showing that students median GPAs during their four years
of study were probably different.
The present conclusion using the Friedman’s Two-Way Analysis of Variance by ranks does not however adjust for
and reflect any possible gradient in the proportions of students experiencing increase (decrease) in their GPAs during the
four years of study. There are observable progressive increases and decreases in these proportions for the present data,
making the application of the Friedman’s Two-Way Analysis of Variance test by rank rather inappropriate here. In fact the
calculated chi-square values of 42.50 for the proposed method and only 25.14 for the Friedman’s Two-Way Analysis of
Variance test by ranks indicate at least for the present data that it is likely to accept a false null hypothesis (Type II error)
more frequently than the proposed method even if there exist no inbuilt ordering in the variables under study.
VI. SUMMARY AND CONCLUSION
This paper has presented and discussed a nonparametric statistical method for the analysis of related measures that
are ordered in time, space or condition. The proposed method may be used to check for the existence of any gradient in
proportions of responses and whether subjects are progressively experiencing increase (or decrease) in their performance
levels. The method is illustrated with some data and shown to be more powerful than the Friedman’s Two-Way Analysis of
Variance test by ranks especially when the data being analyzed have inbuilt order.
REFERENCES
[1]. Bartholomew, D. J. (1959): A test of homogeneity for ordered alternatives. Biometrika, 46: 36-48
[2]. Bartholomew, D. J. (1963): On Chassan’s test for order. Biometrika, 19, 188-191
[3]. Cochran, W. G. (1950): “The Comparison of Percentages in Matched Samples.” Biometrika, 37, 256-266
[4]. Cohen, J. (1983): The Cost of Dichotomization, Appl. Pschol. Measure, 7: 249-253
[5]. Gart, J. J. (1963): A Median test with Sequential Application. Biometrika, 50: 55-62
[6]. Gibbons, J. D. (1971): Nonparametric Statistical Inference, McGraw Hill, New York
[7]. Miller, J (1996): Statistics for Advanced level (2nd
Edition) Cambridge University Press
[8]. Oyeka, C. A. (2010): An Introduction to Applied Statistical Methods (8th
Edition). Nobern Avocation Publishing Company, Enugu,
Nigeria
[9]. Oyeka, C. A.; Utazi C. E.; Nwosu, C. R.; Ebuh, G. U.; Ikpegbu, P. A.; Obiora-Ilouno H. O.; and Nwankwo, C. C. (2010): Non-
Homogenous Sample Data. International Journal of Mathematics Research Vol. 2, No 1, Pp 93-99](https://image.slidesharecdn.com/b03506011-150911111519-lva1-app6891/85/Ties-Adjusted-Nonparametric-Statististical-Method-For-The-Analysis-Of-Ordered-C-Repeated-Measures-6-320.jpg)
Ties Adjusted Nonparametric Statististical Method For The Analysis Of Ordered C Repeated Measures
- 1. International Journal of Mathematics and Statistics Invention (IJMSI) E-ISSN: 2321 – 4767 P-ISSN: 2321 - 4759 www.ijmsi.org Volume 3 Issue 5 || July. 2015 || PP-06-11 www.ijmsi.org 6 | Page Ties Adjusted Nonparametric Statististical Method For The Analysis Of Ordered C Repeated Measures 1 Matthew M. C.; 2 Adigwe, R. U.; 3 Oyeka, I. C. A. and 4 Ajibade Bright 1-2 Department of Mathematics and Statistics, Delta State Polytechnic, Ogwashi-Uku. 3 Department of Statistics, Faculty of Physical Sciences, Nnamdi Azikiwe University, PMB 5025, Awka, Anambra State. 4 Department of General Studies, Petroleum Training Institute, P.M.B 20, Effurun, Warri. Delta State. Nigeria. ABSTRACT : This paper proposes a nonparametric statistical method for the analysis of repeated measures that adjusts for the possibility of tied observation in the data which may be observations in time, space or condition. The proposed method develops a test statistic for testing whether subjects are progressively improving (or getting worse) in their experience over time, space or condition. The method is illustrated with some data and shown to be at least as powerful as the Friedman’s two way analysis of variance test by ranks even in the absence of any in-built ordering in the variable under study. I. INTRODUCTION If one has repeated measures randomly drawn from a number of related populations that are dependent on some demographic factors or conditions or that are ordered in time or space which do not satisfy the necessary assumptions for the use of parametric tests, then use of nonparametric methods is indicated and preferable. These types of data include subjects’ or candidates’ scores in examinations or job placement interviews at various points in time; diagnostic test results repeated a certain number of times; commodity prizes at various times, location or market places. Statistical analysis of these types of data often require the use of nonparametric methods such as the Friedman’s two way analysis of variance test by ranks or the Cochran’s Q- test (Gibbons 1971, Oyeka 2010). However a problem with these two statistical methods is that the Friedman’s test often tries to adjust for ties that occur in blocks or batches of sample observations by assigning these tied observations their mean ranks, the Cochran’s Q-test requires the observations to be dichotomous, assuming only two possible values. Furthermore, if the null hypothesis to be tested is that subjects are increasingly performing better or worse with time or space; then these two statistical procedures may not be readily applicable. In this case the methods developed by Bartholomew and others (Bartholomew 1959 and 1963) may then be available for use. However, some of these methods are rather difficult to apply in practice and the resolution of any ties that may occur within blocks of observations is not often easy. Authors who have worked on these areas include: Oyeka et al (2010), Oyeka (2010), Krauth (2003), Miller (1996), Vargha et al (1996), Cohen (1983), Gart (1963) and Cochran (1950). In this paper we propose a nonparametric statistical method for the analysis of ordered repeated measures that are related in time, space or condition that takes account of all possible pair-wise combinations of treatment levels. II. THE PROPOSED METHOD Let 𝑥𝑖1 𝑥𝑖2 … 𝑥𝑖𝑐 be the 𝑖 𝑡 batch or block in a random sample of 𝑛 observations drawn from some 𝑐 related populations 𝑋1, 𝑋2, … , 𝑋𝑐 for 𝑖 = 1,2, … , 𝑛. where 𝑐 may be indexed in time, space or condition. Population 𝑋1, 𝑋2, … , 𝑋𝑐 may be measures on as low as the ordinal scale and need not be continuous. The problem of research interest here is to determine whether subjects are on the average progressively increasing, experiencing no change or worsening in their score or performance overtime, space or remission of condition. It is quite possible that within any specified time interval say some subjects’ scores at some time in the interval may be higher than their scores earlier in the interval which are themselves higher than their scores later in the interval. To adjust for this possibility we develop ties adjusted extended sign test for this purpose that structurally corrects for ties and also considers all possible pair-wise combinations of treatment levels, we may first take pair-wise combinations of the 𝑐 observations for each of the 𝑛 subjects. Thus suppose for the 𝑖 𝑡 subject the 𝑐′ = 𝑐 2 = 𝑐 𝑐−1 2 pairs of the observations on the 𝑐 treatment levels or population is 𝑥𝑖1, 𝑥𝑖2 , 𝑥𝑖1, 𝑥𝑖3 , … , 𝑥𝑖1, 𝑥𝑖𝑐 , 𝑥𝑖2, 𝑥𝑖3 , 𝑥𝑖2, 𝑥𝑖4 , … , 𝑥𝑖𝑐−1, 𝑥𝑖𝑐 which are increasingly (decreasingly) indexed in time or space, for 𝑖 = 1,2, … , 𝑛. Let
- 2. Ties Adjusted Nonparametric Statististical... www.ijmsi.org 7 | Page 𝑈𝑖𝑗 = 1 𝑖𝑓 𝑡𝑒 𝑓𝑖𝑟𝑠𝑡 𝑜𝑏𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛 𝑖𝑠 𝑖𝑔𝑒𝑟 𝑏𝑒𝑡𝑡𝑒𝑟, 𝑚𝑜𝑟𝑒 𝑡𝑎𝑛 𝑡𝑒 𝑠𝑒𝑐𝑜𝑛𝑑 𝑜𝑏𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛 𝑖𝑛 𝑡𝑒 𝑗 𝑡 𝑝𝑎𝑖𝑟 𝑜𝑓 𝑜𝑏𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛𝑠 𝑓𝑜𝑟 𝑡𝑒 𝑖 𝑡 𝑠𝑢𝑏𝑗𝑒𝑐𝑡 0, 𝑖𝑓 𝑡𝑒 𝑓𝑖𝑟𝑠𝑡 𝑎𝑛𝑑 𝑠𝑒𝑐𝑜𝑛𝑑 𝑜𝑏𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑠 𝑖𝑛 𝑡𝑒 𝑗 𝑡 𝑝𝑎𝑖𝑟 𝑜𝑓 𝑜𝑏𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛𝑠 𝑓𝑜𝑟 𝑡𝑒 𝑖 𝑡 𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑎𝑟𝑒 𝑡𝑒 𝑠𝑎𝑚𝑒 −1, 𝑖𝑓 𝑡𝑒 𝑓𝑖𝑟𝑠𝑡 𝑜𝑏𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛 𝑖𝑠 𝑙𝑜𝑤𝑒𝑟, 𝑤𝑜𝑟𝑠𝑒, 𝑠𝑚𝑎𝑙𝑙𝑒𝑟 𝑡𝑎𝑛 𝑡𝑒 𝑠𝑒𝑐𝑜𝑛𝑑 𝑜𝑏𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛 𝑖𝑛 𝑡𝑒 𝑗 𝑡 𝑝𝑎𝑖𝑟 𝑜𝑓 𝑜𝑏𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛𝑠 𝑓𝑜𝑟 𝑡𝑒 𝑗 𝑡 𝑠𝑢𝑏𝑗𝑒𝑐𝑡 − − − (1) For 𝑖 = 1,2, … 𝑛; 𝑗 = 1,2, … , 𝑐′ = 𝑐 𝑐−1 2 Note that by its specification, 𝑈𝑖𝑗 for 𝑖 = 1,2, … 𝑛; 𝑗 = 1,2, … , 𝑐′ = 𝑐 𝑐−1 2 spans all the 𝑐′ = 𝑐 2 pairwise combinations of the 𝑐 treatment levels. Now let 𝜋𝑗 + = 𝑃 𝑈𝑖𝑗 = 1 ; 𝜋𝑗 0 = 𝑃 𝑈𝑖𝑗 = 0 ; 𝜋𝑗 − = 𝑃 𝑈𝑖𝑗 = −1 − − − − − − − − − (2) Where 𝜋𝑗 + + 𝜋𝑗 0 + 𝜋𝑗 − = 1 − − − − − − − − − − − − − − − − − (3) Note that 𝜋𝑗 0 provides an adjustment for any possible tied scores or observations by subjects for the 𝑗 𝑡 pair of treatment combinations. Let 𝑊𝑗 = 𝑈𝑖𝑗 𝑛 𝑖=1 − − − − − − − −(4) and 𝑊 = 𝑊𝑗 𝑐′ 𝑗 =1 = 𝑈𝑖𝑗 𝑛 𝑖=1 𝑐′ 𝑖=1 − − − − − − − − − − − (5) Now 𝐸 𝑈𝑖𝑗 = 𝜋𝑗 + − 𝜋𝑗 − ; 𝑉𝑎𝑟 𝑈𝑖𝑗 = 𝜋𝑗 + + 𝜋𝑗 − − 𝜋𝑗 + − 𝜋𝑗 − 2 − − − −(6) Also 𝐸 𝑊𝑗 = 𝐸 𝑈𝑖𝑗 𝑛 𝑖=1 = 𝑛 𝜋𝑗 + − 𝜋𝑗 − − − − − − − − − − −(7) And 𝑉𝑎𝑟 𝑊𝑗 = 𝑉𝑎𝑟 𝑈𝑖𝑗 𝑛 𝑖=1 = 𝑛 𝜋𝑗 + + 𝜋𝑗 − − 𝜋𝑗 + − 𝜋𝑗 − 2 − − − − 8 Similarly, 𝐸 𝑊 = 𝐸 𝑊𝑗 𝑐′ 𝑗 =1 = 𝑛 𝜋𝑗 + − 𝜋𝑗 − 𝑐′ 𝑗 =1 − − − −(9) And 𝑉𝑎𝑟 𝑊 = 𝑛 𝜋𝑗 + + 𝜋𝑗 − − 𝜋𝑗 + − 𝜋𝑗 − 2 𝑐′ 𝑗 =1 − − − − − − − (10) Note from equation 9 that 𝐸 𝑊 𝑛 = 𝜋𝑗 + − 𝜋𝑗 − 𝑐′ 𝑗 =1 measures the difference between the proportion of subjects who on the average successively or progressively earn higher scores and the proportion of subjects who on the average successively earn lower scores over time, space or condition. Research interest is often in determining whether a population of subjects are on the average progressively experiencing increases (or decreases) in their scores or performance. For example in the case of disease diagnosis research interest may be on whether the proportion of subjects or patients progressively experiencing remission of disease are progressively increasing (or decreasing) over time or condition. In other words, null hypothesis of interest may be 𝐻0: 𝜋1 + − 𝜋1 − = 𝜋2 + − 𝜋2 − = ⋯ 𝜋 𝑐′ + − 𝜋 𝑐′ − = 𝜋+ − 𝜋− = 0 𝑣𝑠 𝐻1: 𝜋𝑗 + − 𝜋𝑗 + > 0 𝑠𝑎𝑦 − (11) For some 𝑗 = 1,2, … , 𝑐′ The null hypothesis of eqn 11 may be tested based on 𝑊 and its variance. We however here adopt an alternative approach based on the chi-square test for independence. Now 𝜋𝑗 + , 𝜋𝑗 0 𝑎𝑛𝑑 𝜋𝑗 − are respectively the probabilities that in the 𝑗 𝑡 pair of observations or scores, the first score by a randomly selected subject is on the average
- 3. Ties Adjusted Nonparametric Statististical... www.ijmsi.org 8 | Page higher (better, more), the same as (equal to) or lower (worse, smaller) than the second score for 𝑗 = 1,2, … , 𝑐′ = 𝑐 𝑐−1 2 . Their sample estimates are respectively 𝜋𝑗 + = 𝑝𝑗 + = 𝑓𝑗 + 𝑛 ; 𝜋𝑗 0 = 𝑝𝑗 0 = 𝑓𝑗 0 𝑛 ; 𝜋𝑗 − = 𝑝𝑗 − = 𝑓𝑗 − 𝑛 … . 12 where 𝑓𝑗 + , 𝑓𝑗 0 𝑎𝑛𝑑 𝑓𝑗 − are respectively the total number of times the first observation or score is higher, equal to, or lower than the second observation or score in the 𝑗 𝑡 pair of observations. That is 𝑓𝑗 + , 𝑓𝑗 0 𝑎𝑛𝑑 𝑓𝑗 − are respectively the number of 1’s, 0’s and -1’s in the frequency distribution of the 𝑛 values of these numbers in 𝑈𝑖𝑗 , 𝑖 = 1,2, … 𝑛 for each 𝑗. Note 𝑓𝑗 0 = 𝑛 − 𝑓𝑗 + − 𝑓𝑗 − ; 𝑝𝑗 0 = 1 − 𝑝𝑗 + − 𝑝𝑗 − − − − − − −(13) Also let 𝑓+ , 𝑓− 𝑎𝑛𝑑 𝑓0 be respectively the total number of 1’s, -1’s and 0’s in all the 𝑐′ pair wise treatment combinations. That is 𝑓+ = 𝑓𝑗 + 𝑐′ 𝑖=1 ; 𝑓− = 𝑓𝑗 − 𝑐′ 𝑖=1 ; 𝑓0 = 𝑓𝑗 0 𝑐′ 𝑖=1 = 𝑛𝑐′ − 𝑓+ − 𝑓− = 𝑛𝑐 𝑐 − 1 2 − 𝑓+ − 𝑓− (14) Note that 𝑓+ , 𝑓− 𝑎𝑛𝑑𝑓0 are respectively the total number of times subjects successively score higher, lower or experience non change in scores for all the pair-wise treatment combinations. The overall sample proportions for all the 𝑐′ pair-wise treatment combinations are respectively 𝜋+ = 𝑃+ = 𝑓+ 𝑛𝑐′ = 2𝑓+ 𝑛𝑐 𝑐 − 1 ; 𝜋− = 𝑃− = 𝑓− 𝑛𝑐′ = 2𝑓− 𝑛𝑐 𝑐 − 1 ; 𝜋0 = 𝑃0 = 𝑓0 𝑛𝑐′ = 2𝑓0 𝑛𝑐 𝑐 − 1 = 1 − 𝑃+ − 𝑃− − − − −(15) Now in terms of contingency tables, the observed frequencies for these three situations for the 𝑗 𝑡 pair of treatment combinations are respectively 𝑂1𝑗 = 𝑓𝑗 + ; 𝑂2𝑗 = 𝑓𝑗 − ; 𝑂3𝑗 = 𝑓𝑗 0 = 𝑛 − 𝑓𝑗 + − 𝑓𝑗 − − − − −(16) The corresponding proportions are given by equation 12. Now under the null hypothesis of equation 11 that subjects are on the average as likely to progressively earn higher as lower score at all treatment levels, then the expected number of 1’s (higher scores), -1’s (lower scores) or 0’s (no change in scores) at the 𝑗 𝑡 pair of treatment combinations are respectively 𝐸1𝑗 = 𝑛𝑓+ 𝑛𝑐′ = 𝑛𝑓+ 𝑛𝑐 𝑐 − 1 2 ; 𝐸2𝑗 = 𝑛𝑓− 𝑛𝑐′ = 𝑓− 𝑐′ ; 𝐸3𝑗 = 𝑛𝑓0 𝑛𝑐′ = 𝑓0 𝑐′ = 𝑛𝑐′ − 𝑓+ − 𝑓0 𝑐′ (17) Under the null hypothesis of equation 11 of no difference in response calls, that is of equal population medians, the test statistic 𝜒2 = 𝑂𝑖𝑗 − 𝐸𝑖𝑗 2 𝐸𝑖𝑗 𝑐′ 𝑗 =1 3 𝑖=1 − − − − − − − (18) has approximately the chi-square distribution with 3 − 1 𝑐′ − 1 = 2 𝑐′ − 1 = 𝑐 𝑐 − 1 − 2 degrees of freedom for sufficiently large 𝑛. Using equations 16 and 17 in equation 18 we have that 𝜒2 = 𝑓𝑗 + − 𝑓+ 𝑐′ 2 𝑓+ 𝑐′ 𝑐′ 𝑗 =1 + 𝑓𝑗 − − 𝑓− 𝑐′ 2 𝑓− 𝑐′ 𝑐′ 𝑗 =1 + 𝑛 − 𝑓+ − 𝑓− − 𝑛 𝑛𝑐′ − 𝑓+ − 𝑓− 𝑐′ 2 𝑛 𝑛𝑐′ − 𝑓+ − 𝑓− 𝑛𝑐′ 𝑐′ 𝑗 =1 = 𝑐′ 𝑓+ 𝑓− 𝑛𝑐 − 𝑓+ − 𝑓− 𝑓− 𝑛𝑐′ − 𝑓+ − 𝑓− 𝑓+ − 𝑓+ 𝑐′ 2𝑐′ 𝑗 =1 + 𝑓+ 𝑛𝑐′ − 𝑓+ − 𝑓− 𝑓𝑗 − − 𝑓− 𝑐′ 2 + 𝑓+ 𝑓− 𝑓𝑗 + − 𝑓+ 𝑐′ + 𝑓𝑗 − − 𝑓− 𝑐′ 2𝑐′ 𝑗 =1 𝑐′ 𝑗=1 which when further simplified reduces to 𝜒2 = 𝑐′ 𝑓+ 𝑓− 𝑛𝑐′ − 𝑓+ − 𝑓− 𝑓− 𝑛𝑐′ − 𝑓− 𝑓𝑗 + − 𝑓+ 𝑐′ 2𝑐′ 𝑗 =1 + 𝑓+ 𝑛𝑐′ − 𝑓+ 𝑓𝑗 − − 𝑓− 𝑐′ 2 + 2𝑓+ 𝑓− 𝑓𝑗 + − 𝑓+ 𝑐′ 𝑓𝑗 − − 𝑓− 𝑐′ 𝑐′ 𝑗 =1 𝑐′ 𝑗 =1 19
- 4. Ties Adjusted Nonparametric Statististical... www.ijmsi.org 9 | Page which has a chi-square distribution with 2 𝑐′ − 1 = 𝑐 𝑐 − 1 − 2 degrees of freedom; and may be used to test the null hypothesis of equation (11) 𝐻0 is rejected at the 𝛼 −level of significance if 𝜒2 ≥ 𝜒2 1−𝛼,𝑐 𝑐−1 −2 − − − − − 20 otherwise 𝐻0 is accepted. Now an alternative expression for the statistic of equation (19) in terms of the sample proportions of equations 12 − 15 when simplified becomes 𝜒2 = 𝑛 𝑝+ 𝑝− 1−𝑝+−𝑝− 𝑝− 1 − 𝑝− 𝑝𝑗 + − 𝑝+ 2 + 𝑝+ 1 − 𝑝+ 𝑝𝑗 − − 𝑝− 2 + 2𝑝+ 𝑝− 𝑝𝑗 + −𝑐′ 𝑗=1 𝑐′ 𝑗=1 𝑐′ 𝑗=1 𝑝+𝑝−− 𝑝−−−(21) An easier to use computational form of equation (21) is 𝜒2 = 𝑛 𝑝+ 𝑝− 1 − 𝑝+ − 𝑝− 𝑝− 1 − 𝑝− 𝑝𝑗 +2 − 𝑐′ 𝑝+2 𝑐′ 𝑗 =1 + 𝑝+ 1 − 𝑝+ 𝑝𝑗 −2 − 𝑐′ 𝑝−2 𝑐′ 𝑗 =1 + 2𝑝+ 𝑝− 𝑝𝑗 + 𝑝− − 𝑐′ 𝑝+ 𝑝− 𝑐′ 𝑗 =1 (22) which also has a chi-square distribution with 2 𝑐′ − 1 = 𝑐 𝑐 − 1 − 2 degrees of freedom, 𝐻0 is rejected at the 𝛼 −level of significance if equation 20 is satisfied otherwise 𝐻0 is accepted. V. ILLUSTRATIVE EXAMPLE We here use the data of table 1 on the grade point average (GPA) of a random sample of 17 students during each of their four years of study for a degree in a programme of a certain University to illustrate the proposed method. The data are presented in table 1 which also shows the pair-wise difference 𝑑𝑖𝑗 between the GPA’s for 4 years for 𝑖 = 1,2, … ,6. These six pair-wise differences, 𝑑𝑖𝑗 , are taken and used here for simplicity of presentation because the data being analyzed are numeric. However the analysis could still be done by comparing the GPA’s in pairs for each subject over the four years and applying equation 1. TABLE 1: DATA ON GRADE POINT AVERAGE (GPA) FOR FOUR YEARS FOR A RANDOM SAMPLE OF STUDENTS S/No Year 1 Year 2 Year 3 Year 4 𝑑𝑖12 𝑑𝑖13 𝑑𝑖14 𝑑𝑖23 𝑑𝑖24 𝑑𝑖34 1 3.7 1.7 2.2 4.0 2.0 1.5 -0.3 -0.5 -2.3 -1.8 2 3.8 3.3 4.4 4.6 0.5 -0.6 -0.8 -1.1 -1.3 -0.2 3 4.1 4.0 4.4 4.3 0.1 -0.3 -0.2 -0.4 -0.3 0.1 4 4.2 3.1 2.5 3.8 1.1 1.7 0.4 0.6 -0.7 -1.3 5 3.7 3.3 4.3 4.3 0.4 -0.6 -0.6 -0.1 -1.0 0.0 6 3.7 2.9 4.1 3.6 0.8 -0.4 0.1 -1.2 -0.7 0.5 7 2.8 2.1 3.1 3.3 0.7 -0.3 -0.5 -1.0 -1.2 -0.2 8 3.7 2.9 2.8 4.0 0.8 0.9 -0.3 0.1 -1.1 -1.2 9 4.1 2.7 4.0 3.9 1.4 0.1 0.2 -1.3 -1.2 0.1 10 3.0 2.8 2.6 4.0 0.2 0.4 -1.0 0.2 -1.2 -1.4 11 3.5 2.5 3.7 3.7 1.0 -0.2 -0.2 -1.2 -1.2 0.0 12 3.5 3.1 4.0 3.9 0.4 -0.5 -0.4 -0.9 -0.8 0.1 13 4.5 4.4 4.6 4.7 0.1 -0.1 -0.2 -0.2 -0.3 -0.1 14 4.0 3.4 4.3 4.2 0.6 -0.3 -0.2 -0.9 -0.8 0.1 15 3.8 3.5 3.9 4.0 0.3 -0.1 -0.2 -0.4 -0.5 -0.1 16 3.4 3.0 4.0 4.6 0.4 -0.6 -1.2 -1.0 -1.6 -0.6 17 3.9 4.0 4.4 4.7 -0.1 -0.5 -0.8 -0.4 -0.7 -0.3
- 5. Ties Adjusted Nonparametric Statististical... www.ijmsi.org 10 | Page Now to illustrate the proposed method we apply equation 2 to the differences 𝑑𝑖𝑗 of table 1 to obtain values of 𝑈𝑖𝑗 shown in table 2. TABLE 2: VALUES OF 𝑼𝒊𝒋 (EQN 1) FOR THE DIFFERENCES 𝒅𝒊𝒋 IN TABLE 1 AND OTHER STATISTICS S/No 𝑈𝑖1 𝑈𝑖2 𝑈𝑖3 𝑈𝑖4 𝑈𝑖5 𝑈𝑖6 Total 1 1 1 -1 -1 -1 -1 2 1 -1 -1 -1 -1 -1 3 1 -1 -1 -1 -1 1 4 1 1 1 1 -1 -1 5 1 -1 -1 -1 -1 0 6 1 -1 1 -1 -1 1 7 1 -1 -1 -1 -1 -1 8 1 1 -1 1 -1 -1 9 1 1 1 -1 -1 1 10 1 1 -1 1 -1 -1 11 1 -1 -1 -1 -1 0 12 1 -1 -1 -1 -1 1 13 1 -1 -1 -1 -1 -1 14 1 -1 -1 -1 -1 1 15 1 -1 -1 -1 -1 -1 16 1 -1 -1 -1 -1 -1 17 -1 -1 -1 -1 -1 -1 𝑛 17 17 17 17 17 17 102 𝑛𝑐 𝑐−1 2 𝑓𝑗 + 16 5 3 3 0 5 32 𝑓+ 𝑓𝑗 − 1 12 14 14 17 10 68 = 𝑓+ 𝑓𝑗 0 0 0 0 0 0 2 2 = 𝑓0 𝑝𝑗 + 0.941 0.294 0.176 0.176 0.00 0.294 0.314 (= 𝑃+ ) 𝑝𝑗 − 0.059 0.706 0.824 0.824 1.00 0.588 0.667 (= 𝑃−) 𝑝𝑗 0 0.00 0.00 0.00 0.00 0.00 0.118 0.02= 𝑃0 Using the sample proportion shown at the bottom of table 2 in the computational formular of equation 22 we have 𝜒2 = 17 0.667 0.333 1.119 − 6 0.314 2 + 0.314 0.686 3.205 − 6 0.667 2 0.314 0.667 0.020 = 17 0.222 0.527 + 0.215 0.536 + 0.419 −0.530 0.004 = 17 0.117+0.115−0.222 0.004 = 17 0.010 0.004 = 42.50 𝑝 − 𝑣𝑎𝑙𝑢𝑒 = 0.000 which with 𝑐 𝑐 − 1 − 2 = 4 3 − 2 = 12 − 2 = 10 degrees of freedom is highly statistically significant, indicating that students did not progressively earn higher GPAs during their years of study. One may wish to compare the present results with what could have been obtained if the Friedman’s Two-Way Analysis of variance Test by ranks had been used to analyze the data. To do this we may rank the GPAs of each student in table 1 from the lowest assigned the rank 1 to the highest assigned the rank 4. Tied GPAs for each student are assigned their mean ranks. The results are presented in table 3.
- 6. Ties Adjusted Nonparametric Statististical... www.ijmsi.org 11 | Page TABLE 3: Ranks assigned to students GPAs of Table 1 for use with the Friedman’s Test. S/No Year 1 Year 2 Year 3 Year 4 1 3 1 2 4 2 2 1 3 4 3 2 1 4 3 4 4 2 1 3 5 2 1 3.5 3.5 6 3 1 4 2 7 2 1 3 4 8 3 2 1 4 9 4 1 3 2 10 3 2 1 4 11 2 1 3.5 3.5 12 2 1 4 3 13 2 1 3 4 14 2 1 4 3 15 2 1 3 4 16 2 1 3 4 17 1 2 3 4 𝑅.𝑗 41 21 49 59 Now using the sum of ranks given at the bottom of table 3 with the Friedman’s Two-Way Analysis of Variance by Ranks test statistic given as 𝜒2 = 12 𝑛𝑐 𝑐 + 1 𝑅.𝑗 2 4 𝑗 =1 − 3𝑛 𝑐 + 1 − − − − − − − −(23) we have 𝜒2 = 12 412 + 212 + 492 + 592 17 4 5 − 3 17 5 = 12 1681 + 441 + 2401 + 3481 340 − 255 = 280.14 − 255 = 25.14 (𝑝 − 𝑣𝑎𝑙𝑢𝑒 = 0.000) which with 3 degrees of freedom is also statistically significant showing that students median GPAs during their four years of study were probably different. The present conclusion using the Friedman’s Two-Way Analysis of Variance by ranks does not however adjust for and reflect any possible gradient in the proportions of students experiencing increase (decrease) in their GPAs during the four years of study. There are observable progressive increases and decreases in these proportions for the present data, making the application of the Friedman’s Two-Way Analysis of Variance test by rank rather inappropriate here. In fact the calculated chi-square values of 42.50 for the proposed method and only 25.14 for the Friedman’s Two-Way Analysis of Variance test by ranks indicate at least for the present data that it is likely to accept a false null hypothesis (Type II error) more frequently than the proposed method even if there exist no inbuilt ordering in the variables under study. VI. SUMMARY AND CONCLUSION This paper has presented and discussed a nonparametric statistical method for the analysis of related measures that are ordered in time, space or condition. The proposed method may be used to check for the existence of any gradient in proportions of responses and whether subjects are progressively experiencing increase (or decrease) in their performance levels. The method is illustrated with some data and shown to be more powerful than the Friedman’s Two-Way Analysis of Variance test by ranks especially when the data being analyzed have inbuilt order. REFERENCES [1]. Bartholomew, D. J. (1959): A test of homogeneity for ordered alternatives. Biometrika, 46: 36-48 [2]. Bartholomew, D. J. (1963): On Chassan’s test for order. Biometrika, 19, 188-191 [3]. Cochran, W. G. (1950): “The Comparison of Percentages in Matched Samples.” Biometrika, 37, 256-266 [4]. Cohen, J. (1983): The Cost of Dichotomization, Appl. Pschol. Measure, 7: 249-253 [5]. Gart, J. J. (1963): A Median test with Sequential Application. Biometrika, 50: 55-62 [6]. Gibbons, J. D. (1971): Nonparametric Statistical Inference, McGraw Hill, New York [7]. Miller, J (1996): Statistics for Advanced level (2nd Edition) Cambridge University Press [8]. Oyeka, C. A. (2010): An Introduction to Applied Statistical Methods (8th Edition). Nobern Avocation Publishing Company, Enugu, Nigeria [9]. Oyeka, C. A.; Utazi C. E.; Nwosu, C. R.; Ebuh, G. U.; Ikpegbu, P. A.; Obiora-Ilouno H. O.; and Nwankwo, C. C. (2010): Non- Homogenous Sample Data. International Journal of Mathematics Research Vol. 2, No 1, Pp 93-99