Research Methods 6-10
L6: Data Visualisation
Plotting data
Anscombe’s Quartet
Datasaurus Dozen
How do we interpret graphs?
How to best plot data?
Different ways to plot the same data
Summary
L7: Non-parametric Tests
Non-parametric tests
Interval scales and skews
Mann-Whitney U test
SPSS
Wilcoxon related-samples test
SPSS
Kruskal-Wallis H test
SPSS
Kruskal-Wallis: testing differences
Friedman test
SPSS
Friedman test: testing differences
L8: Resampling
What is resampling?
Motivation
Why use resampling techniques?
Why are resampling approached not more popular?
Outline
Resampling for hypothesis testing
Permutation Tests
Between-subjects randomisation tests
Process
Generalisation
Within-subjects randomisation test
Process
Number of participants
Bootstrap Resamples
Calculating SEM
Calculating a confidence interval
Research Methods 6-10 1
, A one-sample test
Bootstrapping with a model fit
Bootstrap advantages
Other approaches
Issues and concerns
Summary
L9: Bayesian
Bayes Theorem
Posteriors, likelihood and priors
The impact of priors
Bayesian inferential tests
Null hypothesis significance testing
Bayes Factors
Evidence for H0?
Frequentists
Credible intervals
Comparison
L6: Data Visualisation
Plotting data
☁️ Visual explanation compliments statistics and is essential to
understanding data
Anscombe’s Quartet
Anscombe (1973)
Represents that you can have the same summary statistics with different
unerlying data
Anscombe suggested 4 datasets where in all cases:
mean(x) = 9.000; sd(x) = 3.317
mean(y) = 7.500; sd(y) = 2.031
corr(x,y) = 0.816
Research Methods 6-10 2
, The effectiveness of this quartet comes from the fact that four clearly
different and visually distinct datasets are producing the same statistical
properties
Datasaurus Dozen
Matejka & Fitzmaurice, 2017
Represents datasets which are identical over a range of statistical
properties, yet produce dissimilar graphics.
Research Methods 6-10 3
, While the data exhibits normal-seeming statistics, plotting the data reveals
a picture of a dinosaur.
Slightly changing the statistics leads to a variety of different shapes
How do we interpret graphs?
1. Are the measurements clear?
2. Can comparisons between categories be made?
Examples
How to best plot data?
There are many kinds which are each suitable for different situations
Research Methods 6-10 4
L6: Data Visualisation
Plotting data
Anscombe’s Quartet
Datasaurus Dozen
How do we interpret graphs?
How to best plot data?
Different ways to plot the same data
Summary
L7: Non-parametric Tests
Non-parametric tests
Interval scales and skews
Mann-Whitney U test
SPSS
Wilcoxon related-samples test
SPSS
Kruskal-Wallis H test
SPSS
Kruskal-Wallis: testing differences
Friedman test
SPSS
Friedman test: testing differences
L8: Resampling
What is resampling?
Motivation
Why use resampling techniques?
Why are resampling approached not more popular?
Outline
Resampling for hypothesis testing
Permutation Tests
Between-subjects randomisation tests
Process
Generalisation
Within-subjects randomisation test
Process
Number of participants
Bootstrap Resamples
Calculating SEM
Calculating a confidence interval
Research Methods 6-10 1
, A one-sample test
Bootstrapping with a model fit
Bootstrap advantages
Other approaches
Issues and concerns
Summary
L9: Bayesian
Bayes Theorem
Posteriors, likelihood and priors
The impact of priors
Bayesian inferential tests
Null hypothesis significance testing
Bayes Factors
Evidence for H0?
Frequentists
Credible intervals
Comparison
L6: Data Visualisation
Plotting data
☁️ Visual explanation compliments statistics and is essential to
understanding data
Anscombe’s Quartet
Anscombe (1973)
Represents that you can have the same summary statistics with different
unerlying data
Anscombe suggested 4 datasets where in all cases:
mean(x) = 9.000; sd(x) = 3.317
mean(y) = 7.500; sd(y) = 2.031
corr(x,y) = 0.816
Research Methods 6-10 2
, The effectiveness of this quartet comes from the fact that four clearly
different and visually distinct datasets are producing the same statistical
properties
Datasaurus Dozen
Matejka & Fitzmaurice, 2017
Represents datasets which are identical over a range of statistical
properties, yet produce dissimilar graphics.
Research Methods 6-10 3
, While the data exhibits normal-seeming statistics, plotting the data reveals
a picture of a dinosaur.
Slightly changing the statistics leads to a variety of different shapes
How do we interpret graphs?
1. Are the measurements clear?
2. Can comparisons between categories be made?
Examples
How to best plot data?
There are many kinds which are each suitable for different situations
Research Methods 6-10 4
