Comparison of Some Parametric and Non-Parametric Statistical Methods

Comparison of Some Parametric and Non-Parametric Statistical Methods

Chapter One

Purpose of the Study

To find out if there exists any relationship between indexes of human development reports using both parametric and nonparametric
To equally test for independence using both multivariate parametric method and nonparametric method to see if it can produce the same
To find out similarities/differences between the two statistical methods based on the result of the analysis used in this
To analyze statistically, multivariate data by using parametric and nonparametric tests for independence

CHAPTER TWO

LITERATURE REVIEW

INTRODUCTION

Multivariate analysis deals with the observation of more than one variable where there is some inherent interdependence between variables. There is a wide variety of multivariate techniques. The choice of the most appropriate method depends on the type of data, the problem, and the sort of objectives that are envisaged for analysis. The review in this chapter extends from the existing literature by providing both multivariate parametric and nonparametric tests for independence.

Multinormality Theory

Multivariate analysis lays too much interest on the assumption that all random vectors come from multivariate normal distribution. By definition, the probability density function of a normal variable with mean m and variance s² is given by

f (x) = (2ps²) exp – ½ (x-m)(s²)^-1(x-m)

Then the extension to the p-variate is

f (x) = (2p ) 2 å

– 1

2 exp-

1 (x – m )1 -1 (x – m )

The reasons for its (normal distribution) preference in the multivariate case are among others. (Hollander M and Wolfe DA, 1973)

The multivariate normal distribution is entirely defined by its first and second
The multivariate distribution is an easy generalization of its univariate counterpart, and the multivariate analysis runs almost parallel to the corresponding analysis based on univariate
Linear functions of a multinormal vector are themselves univariate normal.
In the case of normal variables, zero correlation implies independence and pairwise independence implies total independence.
Equiprobability contours of the multivariate distribution are simple ellipses, which by a suitable change of coordinates can be made into a circle.
When the original data is not multinormal, one can often appeal to central limit theorems which prove that certain functions such as the sample mean are normal for large

Parametric versus non-parametric statistics in the analysis of randomized trials with non-normally distributed data.

The following ideas are the contributions and conclusions gotten from Andrew J Vickers, Wolfowitz, (1942) Siegel & Castellan, (1988) & Dr Matthew Ellis (2002). It has generally been argued that parametric statistics should not be applied to data with non-normal distributions. Empirical research has demonstrated that Mann-Whitney (Wilcoxon) generally has greater power than the t-test unless data are sampled from the normal. In the case of randomized trials, we are typically interested in how an endpoint, such as blood pressure or pain, changes following treatment. Such trials should be analyzed using ANCOVA, rather than a t-test. The objectives of this study were:

CHAPTER THREE

MATERIALS USED FOR THE STUDY

The data set used in this work, listed as Appendix A, consists of eight (8) indicator variables selected from 38 African Countries (HDR 2005).

These indicators are

HDI Human Development Index
EI Education Index
GDP/Index Gross Domestic Product Index
LEI Life Expectancy Index
HPI-I Human Poverty Index values(%)
PNUIWS Population not Using Improved Water Sources
AIR Adult Illiteracy Rate
PBNS Probability at birth of not surviving to age 40(%).

In the indicators table, countries and areas are ranked in descending order by their human development index, value, or by their human poverty index. The indicators present both both similarities or differences. The group of human development Index includes indicators such as LEI, EI, GDP, HDI and the second group includes HPI, PNUIWS, AIR, PBNS.

The human development index (HDI): This measures ‘the average achievements in a country in three basic dimensions of human development,once the dimension indices have been calculated, then the HDI is a simple average of the three dimensions indices. HDI = ⅓ (Life expectancy Index LEI + Education Index El’ + Gross Domestic Product Index ‘GDP’)

CHAPTER FOUR

ANALYSIS

Hoteling’s-T²-Test

Hoteling T² test of independence

CHAPTER FIVE

SUMMARY AND CONCLUSION

Summary

In this research work multivariate parametric and non parametric tests for independence were performed using multivariate data collected from World Bank on Human Development Report. The null hypothesis (Ho: variables are independent) is rejected for both the selected parametric (Hotelling’s T², Wilks Lambda, Canonical correlation analysis) and non parametric (Friedman, Kendall W. Wilcoxon and Cochran Q) tests.

Conclusion

Multivariate parametric and non parametric are two statistical methods of inference. Multivariate parametric methods depend upon the assumption of a specific distributional form for example an approximate multivariate normal distribution. And data for these method will be interval or ratio scales. On the other hand the multivariate non parametric method is referred to as distribution free method using ordinal or normal data and even interval and ratio data when the distributional assumption is unspecified. The analysis of the data from human development report using both methods yields the results. From the parametric test of independence, Hotelling’s T²,

Wilks Lambda and Canonical correlation analysis the null hypothesis that is variable are independent and is rejected in favour of the alternative. From the nonparametric test, Wilcoxon signed-ranks, Friedman, Kendall and Cochran, the null hypothesis is also rejected at the same levels of significant so both methods yield the same result. But a multivariate parametric method for example, Canonical correlation analysis have some times an advantage over nonparametric method. That is, if the null hypothesis is rejected, it shows which of the components hypothesis led to the rejection of the null hypothesis. As well as knowing that the null hypothesis should be rejected, one could enquire which specific linear combination led to its rejection. This work has equally shown us the similarities and differences between parametric and non parametric method in analyzing specific data.

Recommendation

The essence of this work is to expose the relationship between component of human development reports. The result shows that the education index and poverty index are closely related. Hence, assistance need to be given to Countries with high poverty indices in terms of their educational institutions.

References

Bruce, F. and Arusha, C. (2004), parametric and nonparametric test for RIP among the G7 Nations. School of Economics. Discussion paper (unpublished).
Chatfield, C, and Collins, A. J., (1980), Introduction to multivariate analysis. Chapman and Hall. London. NewYork. Tokyo. Melbourne. Madras.
Chernoff, S. and Hofges, L. (1958), results for Wilks’ test of Multivariate independence. Paper pre-printed submitted to Elsevier science (unpublished).
Conover, W. J., (1 999), A practical nonparametric statistics. (Third edition), John Wiley and Sons, New York.
Genest, C. Jean, F.Q. and Bruno, R., (2002) tests of serial independence based on Kendall’s process. The Canadian journal of Statistics vol.30.No.3, page 1-21.
Genest, C and \/erret F. (2003). Locally most powerful rank tests of independencefor copied models (unpublished).
Hallin, M. and Paindaveine, D. (2002), Multivarate signed-ranks: Randles interdirection on Tyler’s Angles (Unpublished).
Harvey, M., (1 995), Intuitive Biostatistics: Choosing a statistical test. Chapter 37. Copyright (c) by Oxford University press Inc.
Horowitz,J.L., (2004), Testing a parametric model against a nonpararnetric alternative with identification through instrumental. Human Development Report 2000. New York.
James, R.S.,(2004), Testing for independence in high dimensions. University of Central Florida, orlando, Florida 328 16- 2370, USA. Jschott(d ucf.edu).
Kai, C.C., et al, (2003), A statistical approach to testing mutual independence of ICA Recovered sources. 4th International Symposium on independent component analysis and Blind signal separation.
Mardia, K.V., Kent, J.T.and Bibby, J.M. (1979), Multivariate analysis. London New York. Toronto Sydney. San Francisco. A subsidiary of Harcourt Brace jovanvich, publishers.

Related Topics:

Other Topics