Data Analysis (Statistical Analysis (Data Reduction (Dimension reduction),…
Sample <10 and/or Normality fail
YES - Non parametric test
ranksum - two sample
signrank - paired test
Kruskal-Wallis - N-sample
Chi-square test - normality
NO - Parametric and Non-parametric
paired t-test - paired test
ANOVA - N-sample
Kolmogorov-smirnov test - normality
t-test - two-sample
Levene's test - variance
rmANOVA - Nsample/repeated
Exact Fisher Test
Linear Classifier (LDA)
Artificial Neural Network
Data Reduction (Dimension reduction)
Backward feature elimination
Decision tree/random forest
Dimension Reduction Link
Multiway array decomposition
Perform Lit Search for how data is normally represented
Is the Raw Data Normally Distributed?
Is this how the data should be represented?
Go to statistical analysis
Use parametric plot + scatter plots
Use non parametric plot
Explore data transformation
FIRST OPTION: Reorganize the Data
Find the Source of Variation (Scatterplots and Histograms)
If Outliers are found
Can/Should outliers be removed?
Yes, there are enough subjects in the study for the power analysis
Remove and restart
Were outliers different in any way NOT related to the study?
1 more item...
Normalize data to each subject
1 more item...
If data is skewed
Is there reasonable way to group subjects to explain variation?
Find a relevant ratio for a linear transformation
Test again for normality
SECOND OPTION: Transform the Data
Perform any common normalizations or transformation according to lit search
For skewed data
Positive Skew with non-negative values
Square Root transformation (for counts or data with a power, e.g. area, volume)
Logarithmic transformation (for cumulative, multiplicative effects or exponential data.
Exponential transformation (for log trends e.g. decay)
Power transformation (for power components e.g. area, volume)
For proportions or Percentages
When looking for a change from normalcy, not a specific increase or decrease
Plot Data for exploration
Non parametric plot
Box-whisker plot - group spread/outliers
Scatter plot - data spread/individual
Histogram - Distribution
Mean + S.E.M. - group difference/spread
Ways to Look at and Interpret Data
Group Subjects by:
Do males and females perform differently either befofe and/or after treatment?
How does recovery vary across time based on treatment? Were there any differences between groups before treatment?
In each group, do subjects show a change over time?