Inferential Statistics: Analysing multiple data sets

1
Inferential Statistics Analysing multiple data
sets

• Ben Cambery - UL009
• bcambery_at_lincoln.ac.uk

2
What weve done before

• Weve looked at parametric and non-parametric
  inferential statistics in relation to comparing
  two groups
• Today were going to consider testing for three
  or more at the same time.
• Were NOT going to perform any of these test by
  hand the formulae are too nasty.

3
What were looking at

• One-Way ANOVA Parametric independent
• Kruskal Wallis H non-parametric independent
• Freidman test non-paramteric paired
• Well mention Univariate / multivariate ANOVA
  the parametric paired test, but its tricky to
  implement in SPSS, so not often used

4
ANOVA ANalysis Of VAriance

• This is the one way variant used as described
• This baby is a big, beefy t-test on steroids
• -Niamh Nic Daeid, 2004
• This test is used directly to compare data sets
  that are
• Normally distributed
• In possession of similar standard deviations
• Ideally interval/ratio and continuous.

5
Demonstration

• Found on SPSS under Compare Means
• Complete data set represents the Dependent List
• Test requires an independent grouping variable or
  Factor this is often something like age,
  location or so on.
• Plug in values and away you go!

6
Kruskal Wallis H test

• The simplest test to use if you want to lie with
  data, as nobody knows what it does!
• -Ben Cambery, 2005
• Non-parametric comparison of three or more data
  sets
• Works on pretty much ANY data configuration, save
  Nominal data

7
Another Demonstration

• Found in K independent samples under
  Nonparametric tests
• NOTE K means more than 2
• Place data to be examined in Test variables
• Place grouping variable in Grouping variable
  box, then define range to compare
• Press OK, and voila!

8
Friedman Test

• Named after its creator, the Russian Cosmologist
  Alexander Alexandrovich Friedman (1888 1925)
  (according to Wikipedia.com)
• Nonparametric test for grouped before and after
  data.
• Works on the majority of data sets, although each
  group should have the same number of values in
  the before and after column

9
Yet Another Demo

• Found in Nonparametric tests, under K related
  samples
• Select groups to compare from list, using the
  Ctrl key to multiple select
• Click OK to run the test

10
A few notes about Univariate / Multivariate ANOVA

• SPSS could handle it better if you find
  yourself needing to play with heavy end ANOVA,
  youre best off using Minitab
• It resides buried deep in the General Linear
  Model tab of SPSS, and requires 2-3 columns of
  grouping variables
• Friedman tests work on parametric data and are
  much simpler to employ

11
Summary

• If your data sets are ordinal, skewed or just
  plain awkward, always side for the non-parametric
  test! It will give you a more reliable answer.

12
Thats all the stats youre getting!

• The exam covers data types, through correlation
  Use of SPSS, questionnaire design and inferential
  tests for 3 groups are NOT EXAMINED!
• Next week Ill be available for a Q and A at this
  time bring your problems and Ill see what I
  can do!
• Anyone with questionnaire issues, Ill be running
  a session at 3pm today in AR1202
• Please keep working through the SPSS booklet it
  will help, honest!