Signposts to help |
Deciding which is the right test to use in a given situation is one of the hardest things to master. Many text books offer Flow charts or 'choice trees' ....Chapter one of SPSS for Windows made Simple by Paul Kinnear and Colin Gray contains 5 that you may find useful. There is a more encompassing chart in Jeremy Foster's book: Data Analysis using SPSS for Windows (versions 8 - 10) on page 21.
For this programme however, these signposts should help. Click the 'Go to Page' for direct access....
Are you trying to find the differences (mean, median etc) between two groups; both measuring the same variable?
Interval
/ Ratio data? Normally distributed? |
Ordinal
data? Non-normal? Ranked? | |
Are
the samples independent of each other, i.e.unpaired, unmatched ? |
Independent sample (student's) 't'-test |
Mann-Whitney 'U'-test for medians |
Are the samples dependent upon each other, matched / paired? |
Paired 't'-test |
|
Is your data only on the Nominal scale?
Then we must simply look for an association between the frequency distributions
because there is no mean or median to find...
>
2 x 2 contingency table? |
2
x 2 contingency table? |
Chi square test | Yates' correction |
If the data set is
small and if three or more factors are involved, the G test becomes more preferable
than Chi
A
test of association for three or more factors? |
The
G- test |
Are you dealing with distributions that have data only on the Ordinal scale?
Do you have just one data set and want to test it against a standard normal distribution
(use 1)
Or do you want to compare two sets to see if they both have the same distribution / come from the same population? (use 2)
1)
Your data against a standard normal distribution? |
2)
Two sets of data against each other |
Kolmogorov-Smirnov one sample test for normality | Two-sample
Kolmogorov-Smirnov test |
Are you trying to assess the strength of the relationship between the values of two variables?
*Use Correlation.....but.....
Is
your data on the Ordinal scale (Non-parametric) |
Is
your data on the Interval or Ratio scale? (Parametric) |
Spearman's rank correlation coefficient test |
Pearson's
product moment correlation coefficient test (PPMCC) |
Are you trying to establish the precise mathematical relationship (i.e. the nature of the relationship) between two variables?
Is the data Parametric?
Is there no more than one dependent variable?
Have you established that the relationship is approximately linear or can be transformed to become linear?
Use Regression:
Are
you interested in estimating the probable value of one variable given a value
of the other? | Are
you working with at least two independent variables and one dependent variable
|
Use
simple Linear regression | Use
multiple regression |
Are you investigating
the differences between three or more samples?
Is there just one dependent variable under investigation?
Use ANOVA's
Does each subject provide just one score?
Use Independent - groups ANOVA's
How many sources of influence are you investigating?
1
source of influence | 2
(or more) sources of influence |
Use
1-factor ANOVA |
Use 2-factor ANOVA's |
Are you trying to deal with two or more dependent variables?
Only
1 dependent variable | 2
(or more) dependent variables |
Use
ANOVA | Use
MANOVA |
see
above |
Does each subject
supply more than one score / reading each i.e. repeated measures?
This can often be measures over time
Use repeated-measure ANOVA's
Do you suspect that there are other variables (must be parametric) that are influencing your results ?
Use ANCOVA
Use
repeated measures ANOVA | Use
ANalysis of COVAriance |
Are
you attempting to reduce a large dataset and at the same time, look for
patterns within the set?
*Consider one of the following multivariate analysis options.
Are you interested in exploring the influence of several variables simultaneously?
Do you already have grouping in existence?
No: use Principal Components Analysis
Yes: use Discriminant Analysis
Do you wish to explore relationships between objects based upon comparisons (similarities and differences) of their measurable or visible features?
Use Cluster Analysis...
To repeat:
No
a priori groupings | a
priori groupings exist | measurable
or visible features available? |
Use
Principal Components Analysis | Use
Discriminant Analysis | Use
Cluster analysis |
* a priori simply means 'in advance', so were there any natural groupings in place before you started or not?
Are you interested
in the spatial distribution of your objects?
[Not included in the Media School pack].......
Do you wish to investigate whether the objects are geographically clustered together, evenly or randomly dispersed? Can you use dendrograms?
Use Nearest Neighbour Analysis
Are you conducting an experiment looking at the distribution of a group of objects (e.g. petrol stations,schools or shoe shops) within an experimental area. Do you wish to know if their distribution is random?
Use Point pattern Analysis.
A quick 'field method' to plot heights (e.g. of landmarks) within an area....
Use Trend surface analysis:
Geographical distribution of objects | Distribution
of objects within an experimental plot |
Mapping
your experimental plot |
Nearest Neighbour analysis | Point pattern analysis | Trend
surface analysis |
Not
included in this pack | Not
included in this pack | N/A |