Page 1 of 1: non parametric statistical test

peach · Credits: 2752 My Scrapbook

statistical tests tht are non parametric
a. regression
b. correlation
c. students test
d. rank correlation
e. none

ans should b e?

BRAVO · Credits: 45300 My Scrapbook

whats the ans peach it should be e

SATVIK · Credits: 958 My Scrapbook

yes it is none of the above

peach · Credits: 2752 My Scrapbook

well the answer given was c] students test

SATVIK · Credits: 958 My Scrapbook

PARAMETRIC TESTS:-
These test hypothesis about the means of variances .
their hypothesis concern interval or ratio scale.
classic examples are - t test (also known as students test- it is a small sample test developed by Gosset as the name 'student', used for samples size <30)
other examples are z test(sample size> 30)
f test,
ANOVA test,
F ratio,
MULTIFACTORIAL ANOVA
POST HOC test

NON PARAMETRIC TEST:-
These tests do not test hypothesis concerning parameters.
These are used to rest nominal/ordinal scale data
These are generally LESS POWERFUL than parametric tests.
Examples :- chi square test

Answer here is none

peach · Credits: 2752 My Scrapbook

ok so its none , thanks

neuron03 · Credits: 3484 My Scrapbook

1st of all u must realize, variables can be classified in this way into- categorical and quantitative. quantitative are the ones where it is possible to obtain an exact numerical value for a measure of central tendency.

now when u r doing a bivariate analysis, in which one variable is categorical and other is quantitative (cat vs. quant bivar) then u have this term- parametric and non-parametric.

the parametric tests in this setting include-
student's t test
paired t test
ANOVA (one wave RCB/ CRD or repeated measures ANOVA)

the non-parametric tests are-
wilcoxon rank sum and sign rank tests (both r v different)
kuskal wallis test
friedman's test
mann whitney u test

these are chosen depending on two things-
1. no. of group/ exposure variable x=/>2
2. distribution of outcome variable y- normal vs. skewed

i m sorry its not possible to explain it more clearly about this, typing here.

one of the best things about AIIMS is that, no matter which course you join, MD, DM , MCh or even MSc, PhD, they train u in research methodology and statistics, which is very important.

mickey_p · Credits: 652 My Scrapbook

Answer is D - rank correlation test

Differentiating between parametric and non parametric tests

I presume you have the basic knowledge of distribution of variables (e.g. Normal/Gaussian distribution), ways of summarizing the distribution – mean, median, mode, std deviation, std error etc. The term STATISTIC refers to mean, median and std deviation of the sample variable whereas the term PARAMETER refers to the mean, median, std error of the same variable for the whole population under study.

Parametric tests work on the underlying premise that we know about the distribution of the variable in the population. This helps us in predicting how a particular statistic will behave in repeated samples of the same size in the same population. E.g. if we take 100 random samples of 50 women in each, from the general population, and compute the mean Hb in each sample, then the distribution of the Hb means across all 100 samples will be approximately like the normal distribution and will be standardized, thus the population mean Hb (parameter) is essentially a ‘mean of means of Hb’ from the 100 random samples (statistic).

Knowing the standardized distribution allows us to compare a particular parameter (e.g. mean systolic BP) across two or more populations. The shape of the distribution may be inverted bell shaped (z or normal distribution for which we use z-tests) or any other (e.g. t-distribution, for which we use t-tests). All t-tests and z-tests are parametric. F test is parametric, so is ANOVA

However in situations in which we do not know about the distribution of the variable in a population we can still make inferences on data using tests that do not rely on the parameter (mean and std error)..hence the term distribution-free or non parametric…

Basically for every parametric test there is a non parametric equivalent (that will be applied in the same sitation as its parametric counterpart except that we do not know the distribution pattern of the variable) e.g. Wilcoxon Signed-Ranks tests is used in a paired design and is the non parametric counterpart for student t-test, and for two independent groups we use the Wilcoxon Rank-Sum Test (or Mann Whitney U) which is the counterpart for independent group t-test). Both these tests compare the equality of medians rather than means (as is done in the parametric tests). Chi square (including Fischer’s test, McNemar test) are non parametric.

Rank correlation test is the non parametric equivalent of Pearson correlation coefficient and is thus non parametric.

As a general rule of thumb – tests that compare means are parametric and tests that compare medians or proportions are non parametric (in situations where the distribution pattern is not know it is better to use median for comparing two groups as mean can fluctuate widely with extreme values whereas median remains pretty much constant even in the presence of high or low values)

Dr Manish

elsa_ned · Credits: 532 My Scrapbook

Sorry, but would someone explain it more the difference between the Chi2-test & the Student test?

		Sign In New User? Sign Up Sign in to access your control panel and messenger!
				TechZone \| SpiderNevi \| HowTo? \| Scrapbook!