David Normando

Download Report

Transcript David Normando

A PowerPoint®-based guide to assist in
choosing the suitable statistical test.
David Normando
NOTE: This presentation has the main purpose to
assist researchers and students in choosing the
appropriate statistical test for studies that
examine one variable (Univariate). Some
multivariates analyses are also included.
Please proceed to the next page ...
If you have any suggestion, criticism, please contact
the author by e-mail: [email protected]
What do you want to do?
For an answer, click on the button
David Normando
1) I want to assess whether my data
have a Normal distribution
2) I want to compare groups
(Looking for differences between samples)
3) I want to make correlation or regression
analysis between variables.
4) I want to check the replicability of data
(analysis of random and systematic error)
5) I would choose the appropriate graph
to my data.
Tests for Data Analysis Distribution- Normality
Normal distribution is requested when using continuos data and n<30
David Normando
You may choose the test according to
sample size.
Use D’Agostino, if n≥10
Use D’Agostino-Pearson, if
n≥20
Use Lilliefors or Shapiro-Wilk,
for any n value
Back to beginning
Comparing groups (samples)
David Normando
What kind of data you have?
(Click on the button).
Parametric
(mean)
NUMERICAL
Continuous
Ex: height / length / weight
(Assuming a normal distribution on n>30)
How to check Normality ?
Ordinal
Nonparametric
Categorical data
Nominal
Ex: Middle (1) / Moderate(2) Severe (3)
Ex: Frequency: Yes / No
Race
Gender
Back to beginning
Numerical Data (parametric)
If the distribution is not Normal,
skip to "Abnormal"
ABNORMAL
David Normando
How many groups (samples) do you have?
1
2
>2
Numerical Data (parametric)
If the distribution is not Normal, skip to "Abnormal"
ABNORMAL
David Normando
Are your samples paired or dependent?
No
Yes
Not sure?
Dependent Samples mean:
Before X After
Left Side X Right Side
T1 x T 2 x T3
Numerical Data (parametric)
If the distribution is not Normal,
skip to "Abnormal"
ABNORMAL
David Normando
Answer: one sample t test
Back to beginning
Numerical Data (parametric)
If the distribution is not Normal,
skip to "Abnormal"
ABNORMAL
David Normando
Answer: Independent t test
or ANOVA.
Back to beginning
Numerical Data (parametric)
If the distribution is not Normal,
skip to "Abnormal"
ABNORMAL
David Normando
Answer: Paired t test or
ANOVA for repeated measurements.
.
Back to beginning
Numerical Data (parametric)
If the distribution is not Normal,
skip to "Abnormal"
ABNORMAL
David Normando
Answer: Analysis of Variance (ANOVA)
or MANOVA (Multiple Analysis of Variance),
if you have >1 variable.
Back to beginning
Ordinal Categorical Data
(Nonparametric)
David Normando
How many groups (samples) do you have ?
2
>2
Ordinal Categorical Data
(Nonparametric)
David Normando
Are your samples paired or dependent?
No
Yes
Not sure?
Dependent Samples mean:
Before X After
Left Side X Right Side
T1 x T 2 x T3
Ordinal Categorical Data
(Nonparametric)
David Normando
Are your samples paired or dependent?
No
Yes
Not sure?
Dependent Samples mean:
Before X After
Left Side X Right Side
T1 x T 2 x T3
Ordinal Categorical Data
(Nonparametric)
David Normando
Answer: Mann-Whitney test
Back to beginning
Ordinal Categorical Data
(Nonparametric)
David Normando
Answer: Wilcoxon (signed rank test)
or Signal test.
Back to beginning
Ordinal Categorical Data
(Nonparametric)
David Normando
Answer: Kruskal-Wallis’ Test
Back to beginning
Ordinal Categorical Data
(Nonparametric)
David Normando
Answer: Friedman’s Test
Back to beginning
Nominal Categorical Data
(Nonparametric)
David Normando
How many groups (samples) do you have ?
2
>2
Nominal Categorical Data
(Nonparametric)
David Normando
Are your samples paired or dependent?
No
Yes
Not sure?
Dependent Samples mean:
Before X After
Left Side X Right Side
T1 x T 2 x T3
Nominal Categorical Data
(Nonparametric)
David Normando
Is there any expected value <5 ?
No
Yes
Not sure?
If some of the cells in the contingency table give
values (expected) lower than 5.
Nominal Categorical Data
(Nonparametric)
David Normando
Are your samples paired or dependent?
No
Yes
Not sure?
Dependent Samples mean:
Before X After
Left Side X Right Side
T1 x T 2 x T3
Nominal Categorical Data
(Nonparametric)
David Normando
Answer: Chi-square (x²) test or
Binomial Test, if using 2 samples and
proportion (%)
Back to beginning
Nominal Categorical Data
(Nonparametric)
David Normando
Answer: Cochran’s test
(absolute or relative frequence: %)
Back to beginning
Nominal Categorical Data
(Nonparametric)
David Normando
Answer: McNemar’s test
Back to beginning
Nominal Categorical Data
(Nonparametric)
David Normando
Answer: Exact Fisher’s test
Back to beginning
Correlation or Regression Analysis
David Normando
What kind of data you have?
(Click on the button).
Parametric
(mean)
Numerical
Ex: height / length / weight
(Assuming a normal distribution)
How to check Normality ?
Ordinals
Ex: Middle (1) / Moderate(2) Severe (3)
Nominal
Ex: Frequency: Yes / No
Race
Gender
Nonparametric
Categorical data
Numerical Data (parametric)
If the distribution is not Normal,
skip to "Abnormal"
ABNORMAL
David Normando
How many variables do you have?
2
>2
Correlation tests or regression
analysis to Continuos data
If the distribution is not Normal, skip to "Abnormal"
ABNORMAL
David Normando
Answer: Pearson’s Correlation
Simple Linear Regression
Back to beginning
Correlation tests or regression
analysis to Continuos data
If the distribution is not Normal, skip to "Abnormal"
ABNORMAL
David Normando
Answer: Pearson’s Correlation (parcial) or
Canonical Correlation
Multiple Linear Regression
NOTE: For Correlation all variables examined must have a Normal Distribution.
For Linear Regression dependent variable must have a Normal Distribution
How to check Normality ?
Back to beginning
Correlation test to Ordinal data
(nonparametric)
David Normando
Answer : Spearman or Kendal Correlation
Back to beginning
Correlation and Regression Analysis
to Nominal data (nonparametric)
David Normando
How many variables do you have?
2
>2
Correlation test to Nominal data
(nonparametric)
David Normando
Answer: Contingency coefficient C
Simple Logistic Regression
Back to beginning
Correlation test to Nominal data
(nonparametric)
David Normando
Answer: Contingency coefficient C
Multiple Logistic Regression
Back to beginning
Replicability or Reproducibility
(Systematic error)
David Normando
What kind of data you have?
(Click on the button).
Parametric
(mean)
Numerical
Ex: height / length / weight
(Assuming a normal distribution)
How to check Normality ?
Ordinal
Nonparametric
Categorical data
Nominal
Ex: Middle (1) / Moderate(2) Severe (3)
Ex: Frequency: Yes / No
Race
Gender
Replicability or Reproducibility
(Systematic error for numerical data)
ABNORMAL
David Normando
Answer: Parametric test for dependent data
2 samples
>2 samples
Note: Intraclass correlation can be used, if
you would like to check the association between 2
or more measurements.
For random or casual error , you may use
TEM (technical error measurement):
D= difference between repeated measures
n=number of individuals
Back to beginning
Replicability or Reproducibility
(Systematic error for ordinal data)
David Normando
Answer: Weighted Kappa
NOTE: in case of an ordinal variable, nonparametric tests
for paired or dependent data can also be used
2 sample
> 2 samples
Back to beginning
Replicability or Reproducibility
(Systematic error for Nominal data)
David Normando
Answer: Kappa
Back to beginning
Graph Selection
David Normando
What kind of data you have?
(Click on the button).
Parametric
(mean)
Nonparametric
Categorical data
Numerical
Ordinal
Nominal
Ex: height / length / weight
(Assuming a normal distribution)
Ex: Middle (1) / Moderate(2) Severe (3)
Ex: Frequency: Yes / No
Race
Gender
Back to beginning
Graph Selection
David Normando
Comparing Independent Samples
Comparing Dependent Samples (paired)
Making Data Correlation or regression
Not sure?
Dependent Samples mean:
Before X After
Left Side X Right Side
T1 x T 2 x T3
BOX-PLOT - Comparing Groups.
Continuous or Ordinal Data (Score)
David Normando
This chart describes the measure of central tendency (MEAN for continuos data or MEDIAN for Ordinal
data), measures of dispersion (Standard deviation for parametric data or interquartiles deviation for
Ordinal data) and the whiskers (maximum and minimum values )
Outlier: an observation that
is numerically distant
from the rest of the data.
Back
Line graph for longitudinal data
David Normando
This chart describes the measure of central tendency (mean for Continuos data or median
for Ordinal data) longitudinally
Back
Graphic for Correlation Tests
or Regression Analysis
David Normando
Back
Bar/Column Graphic
Nominal data (frequency)
David Normando
Back