Topics

rencam (140) emo (120) friends (113) Bokashi (62) makan2 (56) Rozy Tiger (54) family (43) betty (35) Lulu (29) Budak Asrama (28) books (21) sale (20) umah baru (18) Giant (16) Boo Yong (11) momot (11) Mail Lambung (10) PG journey (10) Sunkiss (10) cikgu (9) PETKNODE (8) AIR ASIA SALE (7) holiday (7) spss (7) PhDJourney (6) Synergi (6) movie (5) bake (4) daging (4) pasta (4) Ramadhan (3) ayam (3) sotong (3)

Geng Betty

Tuesday, 1 November 2011

Parametric & Non-parametric Test

As Salam..

Me bz..stil struggling with my SPSS thingy..

Just a short note as a reminder:
  1. always do TEST OF NORMALITY to check ur data distribution..normal or not normally distributed..this is important, it will help u to determine the type of test..either parametric test or non-parametric test..there's  a few item to look at..Shapiro-Wilk (for less than 100 sample) or Kolmogorov-Smirnov (for more than 100 sample), not to forget Q-Q Normal Plot and Boxplot.
  2. For datas which are normally distributed...
  • compare mean with reference/known value - simple student's T-test
  • compare means between 2(only) independent group/sample ie: gender (male & female), diet (vege & non vege) - Independent T-test
  • compare means between 2 dependent group/sample ie: before and 2 repetitive after value/readings/figure/results- Dependent T-test
  • compare means between 3 or more group/sample ie: age range (10-15, 16-20, 21-25, 26-30..total 4 groups) - ANOVA (Analysis of variance) <--there are many types of ANOVA depending on the number of factors involved, one way or 2 way etc..gotta do more reading on this!! and Post-hoc test is needed if there is significant value..just to differentiate which groups/sample is actually giving significant result and which doesn't. Quite a number of post-hoc test..choose one that is suitable..regarding ur cases/sample.
  • useful links for ANOVA:


WHAT IF THE DATA IS NOT NORMALLY DISTRIBUTED??

That would be even easier!! We opt for non-parametric test..instead of comparing means..we compare median instead..so do your descriptive analysis before hand..

  • for 2 different/independent group/sample - Mann Whitney Test
  • for more than 2 group/sample - Kruskal Wallis Test
  • for 2 dependent group/sample - Wilcoxon Test

How about correlation??

Normal distribution - Pearson Correlation
Not normally distributed - Spearman Correlation


P/S: Parametric tests are way better than the non-parametric..so if ur data is not normally distributed...just log(log/Ln) them...


No comments:

Post a Comment

LinkWithin

Related Posts Plugin for WordPress, Blogger...

Usia perkongsian cerita kita...

PitaPata Cat tickers