Upload
drnagunuri-srinivas
View
230
Download
0
Embed Size (px)
Citation preview
7/30/2019 Class8 Non-Parameter Tests
1/49
Non-parametric Tests
Research II MSW PT
Class 8
7/30/2019 Class8 Non-Parameter Tests
2/49
Key Terms
Power of a test refers to the probability of rejectinga false null hypothesis (or detect a relationship whenit exists)
Power Efficiency the power of the test relative tothat of its most powerful alternative. For example,if the power efficiency of a certain nonparametric testfor difference of means with sample size 10 is 0.9, itmeans that if interval scale and the normalityassumptions can be made (more powerful), we canuse the t-test with a sample size of 9 to achieve thesame power.
7/30/2019 Class8 Non-Parameter Tests
3/49
Choice of nonparametric test
It depends on the level of measurement obtained (nominal,ordinal, or interval), the power of the test, whether samples arerelated or independent, number of samples, availability ofsoftware support (e.g. SPSS)
Related samples are usually referred to match-pair (usingrandomization) samples or before-after samples.
Other cases are usually treated as independent samples. Forinstance, in a survey using random sampling, we have a sub-sample of males and a sub-sample of females. They can beconsidered as independent samples as they are all randomly
selected.
7/30/2019 Class8 Non-Parameter Tests
4/49
Level of
measurement
One-sample
test
Two-sample case K-sample case
Related Samples Independent samples Related
samples
Independent
samples
Nominal Binomial McNemar forsignificance of
changes
Fisher exactprobability
Chi-square
Cochran Q(Dichotomous)
Chi-square
Ordinal Kolmogorov
Smirnov
Runs
Sign Wilcoxon
matched-pair
signed-ranks
Mann-Whitney U
Kolmogorov-Smirnov
Wald-Wolfowitz runs
Moses of extremereactions
Friedman
two-way
analysis of
variance
Kendalls W
Kruskal-Wallis
one-way
analysis of
variance
Interval Walsh Randomization
7/30/2019 Class8 Non-Parameter Tests
5/49
One-sample case Binomial tests whether the observed
distribution of dichotomous variable (a
variable that has two values only) is the sameas that expected from a given binomialdistribution.
The default value of p is 0.5.You can changethe value of p.
For example, a couple hasgiven birth consecutively 8 baby girls, and
you would like to test if their probability ofgiven birth to baby girls is > 0.6 or >0.7, youcan test the hypothesis by changing thedefault value of p in the SPSS programme.
7/30/2019 Class8 Non-Parameter Tests
6/49
7/30/2019 Class8 Non-Parameter Tests
7/49
7/30/2019 Class8 Non-Parameter Tests
8/49
Binomial Test
Category N Observed Prop. Test Prop. Exact Sig. (2-tailed)
Group 1 Male 8 1.00 .50 .008Gender
Total 8 1.00
7/30/2019 Class8 Non-Parameter Tests
9/49
Chi-square tests whether the
observed distribution is the same as a
certain hypothesized distribution. The default null hypothesis is even
distribution.
7/30/2019 Class8 Non-Parameter Tests
10/49
7/30/2019 Class8 Non-Parameter Tests
11/49
7/30/2019 Class8 Non-Parameter Tests
12/49
Age
Observed N Expected N Residual
18 - 24 92 55.9 36.1
25 - 34 78 111.8 -33.8
35 - 44 100 167.7 -67.745 - 55 95 111.8 -16.8
56 or above 138 55.9 82.1
Total 503
Test StatisticsAge
Chi-Square(a) 184.003
df 4
Asymp. Sig. .000
a 0 cells (.0%) have expected frequencies less than 5. The minimum expected cell frequency
is 55.9.
7/30/2019 Class8 Non-Parameter Tests
13/49
Kolmogorov-Smirnov Compares
the distribution of a variable with a
uniform, normal, Poisson, orexponential distribution,
Null hypothesis: the observed values
were sampled from a distribution ofthat type.
7/30/2019 Class8 Non-Parameter Tests
14/49
7/30/2019 Class8 Non-Parameter Tests
15/49
7/30/2019 Class8 Non-Parameter Tests
16/49
7/30/2019 Class8 Non-Parameter Tests
17/49
7/30/2019 Class8 Non-Parameter Tests
18/49
Runs
A run is defined as a sequence of cases onthe same side of the cut point. (Anuninterrupted course of some state or
condition, for e.g. a run of good luck). You should use the Runs Test procedure
when you want to test the hypothesis thatthe values of a variable are ordered randomly
with respect to a cut point of your choosing(Default cut point: median.
7/30/2019 Class8 Non-Parameter Tests
19/49
E.g. If you ask 20 students about how well they understand alecture on a scale ranged from 1 to 5 (and the median in theclass is 3). If you find that, the first 10 students give a valuehigher than 3 and the second 10 give a value lower than 3
(there are only 2 runs). 5445444545 2222112211 For random situation, there should be more runs (but will not be
close to 20, which means they are ordered exactly in analternative fashion; for example a value below 3 will be followedby one higher than it and vice versa). 2,4,1,5,1,4,2,5,1,4,2,4
The Runs Test is often used as a precursor to running tests that
compare the means of two or more groups, including: The Independent-Samples T Test procedure. The One-Way ANOVA procedure. The Two-Independent-Samples Tests procedure. The Tests for Several Independent Samples procedure.
7/30/2019 Class8 Non-Parameter Tests
20/49
Note: In this data set, 80 social workers (1) are listed together, and followed by120 non-social workers (2), obviously, the order in not random. Since there aremore non-social workers, the median is still 2. There are only 2 runs, one lowerthan the median (2) and one higher than or equal to it.
7/30/2019 Class8 Non-Parameter Tests
21/49
Runs Test
Social Worker
Test Value(a) 2
Cases < Test Value 80
Cases >= Test Value 120Total Cases 200
Number of Runs 2
Z -14.033
Asymp. Sig. (2-tailed) .000
a Median
7/30/2019 Class8 Non-Parameter Tests
22/49
Sample cases (RelatedSamples)
McNemar tests whether the changesin proportions are the same for pairs of
dichotomous variables. McNemars testis computed like the usual chi-square
test, but only the two cells in which theclassification dont match are used.
Null hypothesis: People are equallylikely to fall into two contradictoryclassification categories.
7/30/2019 Class8 Non-Parameter Tests
23/49
Support New Dawn Project (Before)
Yes No
Yes 19 13Support New Dawn Project (After)No 1 7
7/30/2019 Class8 Non-Parameter Tests
24/49
7/30/2019 Class8 Non-Parameter Tests
25/49
7/30/2019 Class8 Non-Parameter Tests
26/49
Test Statistics (b)
Support New Dawn Project
(Before) & Support New DawnProject (After)
N 40
Exact Sig. (2-tailed) .002(a)a Binomial distribution used.
b McNemar Test
7/30/2019 Class8 Non-Parameter Tests
27/49
Sign test tests whether the numbers ofdifferences (+ve or ve) between twosamples are approximately the same. Eachpair of scores (before and after) arecompared.
When after > before (+ sign), if smaller (-sign). When both are the same, it is a tie.
Sign-test did not use all the informationavailable (the size of difference), but itrequires less assumptions about the sampleand can avoid the influence of the outliers.
7/30/2019 Class8 Non-Parameter Tests
28/49
To test the association between thefollowing two perceptions
Social workers help the disadvantagedand Social workers bring hopes to thosein averse situation
7/30/2019 Class8 Non-Parameter Tests
29/49
7/30/2019 Class8 Non-Parameter Tests
30/49
Frequencies
N
Social workers bring hopes
to those in averse situation- Social workers help thedisadvantaged
Negative
Differences(a) 104
PositiveDifferences(b)
71
Ties(c) 322
Total 497a Social workers bring hopes to those in averse situation < Social workers help the disadvantaged
b Social workers bring hopes to those in averse situation > Social workers help the disadvantagedc Social workers bring hopes to those in averse situation = Social workers help the disadvantaged
7/30/2019 Class8 Non-Parameter Tests
31/49
Test Statistics(a)
Social workers bring hopes to those in aversesituation - Social workers help the
disadvantaged
Z -2.419
Asymp. Sig.
(2-tailed) .016a Sign Test
7/30/2019 Class8 Non-Parameter Tests
32/49
Wilcoxon matched-pairs signed-ranks test Similar to sign test, but take into consideration theranking of the magnitude of the difference among
the pairs of values. (Sign test only considers thedirection of difference but not the magnitude ofdifferences.)
The test requires that the differences (of the truevalues) be a sample from a symmetric distribution
(but not require normality). Its better to run stem-and-leaf plot of the differences.
7/30/2019 Class8 Non-Parameter Tests
33/49
Ranks
N
Mean
Rank
Sum of
RanksNegative Ranks 104(a) 88.07 9159.50
Positive Ranks 71(b) 87.89 6240.50
Ties 322(c)
Social workers bring hopes to those inaverse situation - Social workers helpthe disadvantaged
Total497
a Social workers bring hopes to those in averse situation < Social workers help the disadvantagedb Social workers bring hopes to those in averse situation > Social workers help the disadvantagedc Social workers bring hopes to those in averse situation = Social workers help the disadvantaged
7/30/2019 Class8 Non-Parameter Tests
34/49
Test Statistics (b)
Social workers bring hopes to those in aversesituation - Social workers help the disadvantaged
Z -2.340(a)
Asymp. Sig.(2-tailed)
.019
a Based on positive ranks.b Wilcoxon Signed Ranks Test
7/30/2019 Class8 Non-Parameter Tests
35/49
Two-sample case(independent samples)
Mann-Whitney U similar to Wilcoxon matched-paired signed-ranks test except that the samples areindependent and not paired. Its the most commonlyused alternative to the independent-samples ttest.
Null hypothesis: the population means are the samefor the two groups.
The actual computation of the Mann-Whitney test issimple. You rank the combined data values for thetwo groups. Then you find the average rank in eachgroup.
Requirement: the population variances for the twogroups must be the same, but the shape of thedistribution does not matter.
7/30/2019 Class8 Non-Parameter Tests
36/49
7/30/2019 Class8 Non-Parameter Tests
37/49
Ranks
Sex N Mean Rank Sum of Ranks
Male 229 247.36 56645.50
Female 272 254.06 69105.50
Social Worker
Total 501
Test Statistics (a)
Social Worker
Mann-Whitney U 30310.500
Wilcoxon W 56645.500Z -.628
Asymp. Sig. (2-tailed) .530
a Grouping Variable: Sex
7/30/2019 Class8 Non-Parameter Tests
38/49
Kolmogorov-Smirnov Z to test iftwo distributions are different. It isused when there are only a few valuesavailable on the ordinal scale. K-S testis more powerful than M-W U test if the
two distributions differ in terms ofdispersion instead of central tendency.
7/30/2019 Class8 Non-Parameter Tests
39/49
Test Statistics (a)
SocialWorker
Absolute .036
Positive .009
Most ExtremeDifferences
Negative -.036
Kolmogorov-Smirnov Z .397
Asymp. Sig. (2-tailed) .998a Grouping Variable: Sex
7/30/2019 Class8 Non-Parameter Tests
40/49
Wald-Wolfowitz Run Based on the
number of runs within each group when
the cases are placed in rank order. Moses test of extreme reactions
Tests whether the range (excluding the
lowest 5% and the highest 5%) of anordinal variables is the same in the twogroups.
7/30/2019 Class8 Non-Parameter Tests
41/49
K-sample case(Independent samples)
Kruskal-Wallis One-way ANOVAIts more powerful than Chi-square test
when ordinal scale can be assumed. Itis computed exactly like the Mann-Whitney test, except that there are
more groups. The data must beindependent samples from populationswith the same shape (but notnecessarily normal).
7/30/2019 Class8 Non-Parameter Tests
42/49
7/30/2019 Class8 Non-Parameter Tests
43/49
7/30/2019 Class8 Non-Parameter Tests
44/49
Ranks
Educationlevel N
MeanRank
Primary orlower
105 264.67
Secondary 239 248.41Post-secondary
159 249.03
SocialWorker
Total 503
Test Statistics(a,b)
Social WorkerChi-Square 1.049
df 2
Asymp. Sig. .592a Kruskal Wallis Test
b Grouping Variable: Education level
7/30/2019 Class8 Non-Parameter Tests
45/49
K related samples
Friedman two-way ANOVA test
whether the k related samples could
probably have come from the samepopulation with respect to mean rank.
7/30/2019 Class8 Non-Parameter Tests
46/49
7/30/2019 Class8 Non-Parameter Tests
47/49
7/30/2019 Class8 Non-Parameter Tests
48/49
Ranks
1.88
2.45
1.68
Social Worker
Doctor
Lawyer
Mean Rank
Test Sta t i s t i csa
503
204.2412
.000
N
Chi-Squaredf
Asymp. Sig.
Friedman Testa.
Ranks
1.88
2.45
1.68
Social Worker
Doctor
Lawyer
Mean Rank
Test Stat i s t i c s503
.203
204.241
2
.000
N
Kendall's Wa
Chi-Square
df
Asymp. Sig.
Kendall's Coefficient of Concordancea.
7/30/2019 Class8 Non-Parameter Tests
49/49
Cochran Q determines whether it is
likely that the k related samples could
have come from the same populationwith respect to proportion or frequencyofsuccesses in the various samples.
In other words, it only comparesdichotomous variables.
Lets try this in class