Which Statistical Tests to Use?

Level of
Measurement Statistical Test Requirements/ Limitation(s) Purpose
& Data Type
Chi – square tests for 80% cells need 5 or more expected To test if observed frequencies differ from
Level one, two sample of k cases expected frequencies
Nominal independent samples
(Non – Binomial test for one Can use medians and quartiles. No Tests “yes/no” data for nonrandomness of
Parametric) sample frequency requirements. Use when small, one – sample cases. Useful test for
chi – square requirements not met. small subpopulation.
Type Fisher Exact Use when small N instead of chi – As for chi – square.
Discrete Probability Test square.
Grouped Contingency As for chi – square. To test degrees of association (correlation)
Coefficient between two variables.
Mann – Whitney U Use when t-test requirements not Tests for central tendency, and whether two
Test for two met. Use for small N but not for independent samples from the same population.
independent samples. many tied values. Two samples. Tests for differences and similarities.
Almost as powerful as Needs continuously distributed data.
Level t- test.
Ordinal (NON – Wilcoxon test for two Needs relatively continuous data. As for U-test but for paired samples.
Parametric) related samples. As for U – test but for paired values.
Median Test. Use when U – test has too many tied As for U-test. Uses when U-test requirements
values not met.
Type
Data arranged Signs test. Randomness; independent sets of Tests for differences and significance between
on interval data. Data less normal than two sets of data. Before/ after type situations.
scales which Wilcoxon.
have been given Kolmogorov- Smimov As for median test. Needs relatively Tests similarities of distributions. Alternative
ranks two sample test. continuous data. to chi – square where small sample and many
tied values.
(a) Friedman test Independent random sample (a) two
(b) Kruskal – Wallis samples, (b) k samples Analysis of Variance
test
Level Runs Test Assumes normal distribution To test number of runs in a sequence. To test
Ordinal (Non – number of runs which may be considered
Parametric) random. Tests for differences in dispersion.
Binomial (proportions) Data in percentages or other As for z – scores
Type test proportions
Data arranged Spearman’s rank Cannot handle too many paired To test correlation between variables having
on interval correlation coefficient values. paired values.
scales which Kendall’s rank Can handle many paired values. To test correlation between variables having
have been given correlation paired values.
ranks
Level t-test for two To test match means between samples or
Interval independent samples, groups. Only slightly more powerful than U-
(Parametric) and matched pairs test.
Standard, Z – scores To test probability of an occurrence, or to
Type Approximately normal distribution compare places of individuals on two
Continuous distributions.
Individual Standard Error of To determine confidence limits for samples.
means
Regression To show graphically, relationship between
variables, and likely trends or prediction
Level Analysis of Variance To test significance of differences between
Ratio (ANOVA) means.
(Parametric) Pearson’s Product To test correlation between variables. Very
Type Moment Correlation Approximately normal distribution. Powerful.
Continuous
Individual
 Types of Measurement

Purpose to Measure The Measure to Use Definition

Central tendency/ Mean The average
Central Location/ Median Middle most position
Averages Mode Most frequent values
Proportions Fractions of populations, e.g. 4/5
Percentages Part of 100, e.g. 57%
Relationship in size Ratios Relationship of two numbers, e.g. 1:7
between two numbers Rates Type of ratio for bigger numbers, e.g.
6:100,000, six doctors/100,000 pop.
Range Difference extreme high – low scores
Kelly’s Range (Percentile Difference 10th and 90th percentile
Range)
Dispersion/ Spread or Quartiles Divides a distribution into 4 equal parts
Variability Quartile Deviation Difference between upper (Q3)and lower (Q1)
quartile divided by two
Index of Variability Interquartile range divided by median x 100
Deviation/ Mean Deviation Average differences from mean
Standard Deviation (s.d.) Standardized differences from mean
Variability around Variance S.D. squared
‘average’
Combination of Standard Coefficient of S.D. divided by mean x 100
Central tendency and Variability
spread
Standard or z – score Difference from mean divided by s.d.

LEVELS OF MEASUREMENT AND TYPES OF DATA

Types of Data
Levels of Measurement
Continuous/ Discrete Individual/ Grouped
RATIO
Highest Level CONTINUOUS
Has true zero (can be measured) or Can use INDIVIDUAL
INTERVAL DISCRETE OR GROUPED DATA
Intervals known (can only be counted)
Cannot use + - x -
ORDINAL
Can rank or order but cannot + -
x- Can generally only use
DISCRETE data only
Nominal GROUPED DATA
Symbolic numbers
Lowest level

Types of Quantitative Data

Types Qualitative Quantitative

Primary Own interviews, opinions Own survey; field maps, own
photos
Secondary Reported speeches, oral Survey for some other
histories, diaries, newspaper purpose; published tables;
reports maps; aerial photos