Please enable JavaScript.
Coggle requires JavaScript to display documents.
(Outcome (Continous (Independent observation (2 groups/ time points…
Outcome
Continous
Independent observation
2 groups/ time points
2-sample t-test
Mechanics
Difference in means : T- distribution
(Z distribution for larger samples)
SE of the difference in means (pooled)
More precise estimate of SE
T-distribution has more df
Required Homogeneity of variances
SE of the difference in means (unpooled)
2
F test
Difference in means of the groups more than background noise (=variability within groups) ?
F = (Variability btw groups/ Variability within groups)
Hypothesis : two variances are equal (F statistic =1)
Global test
Total sum of square (TSS)
Sum of Squares within (SSW) or Sum of Squares Error (SSE)
Sum of Squares Between (SSB) or Sum of Squares Regressgion (SSR)
TSS = SSW + SSB
Coefficient of Determination
The amount of variation in the outcome variable (dependent variable) that is explained by the predictor (independent variable).
Corection for multiple comparisons post-hoc
Bonferroni correction
Holm/Hochberg
Tukey (adjusts p)
Scheffe (adjusts p)
Pairwise t-tests?
Type I error (5% each test)
1 - (0.95)3 = 14%
Continous variable
Linear correlation (Pearson): 2 variables treated equals
Covariance
cov(x,y)=0 : X and Y independent
cov(x,y)>0 : X and Y positively correlated
cov(x,y)<0 : X and Y inversely correlated
Var(x) = Cov(x,x)
Correlation coefficient
Strength of linear relationship
0: no correlation (independent)
-1 : perfect inverse correlation
+1: perfect positive correlation
Unitless
PEARSON's Correlation Coefficient
R2 (R-squared)
Proportion of variability explained by the predictors
Measure of model fit
Distribution
Normal for larger n
T-distribution for smaller n (n<100)
Linear regression: Predictor ~ Outcome
Simple linear regression
Y = bx+ a
Intercept
Slope (beta coefficient)
Distribution of beta coefficient
T distribution
Multiple linear regression
Each regression coefficient is the amount of change in the outcome variable that would be expected per one-unit change of the predictor, if all other variables in the model were held constant
Functions
Control for confounders
10% change in adjusted beta
Do not judge confounders by their effect on p-values
Improve predictions
Test of interactions btw predictors (Effect modification)
Categorical predictors?
Binary: Treat as numbers (0 and 1)
Categorical : Dummy coding!
Residual analysis
Residual = Observed - Predicted
Residual analysis for normality
Residual analysis for homogeneity of variances
Residual plots
WATCH OUT
Over fitting
Missing data
Variable transformation
NOT normaly distributed (n<100)
NON-homogenous variances
Predictor and Outcome do not have a linear relationship
Correlated observation
2 groups/ time points
Paired t-test
2
Repeated-measures ANOVA
Question
Are there significant differences across time periods? (Time factor)
Are there significant differences between groups (=your categorical predictor)? (Group factor)
Are there significant differences between groups in their changes over time? (Group x Time) factor
Serial paired t-test : type I error
Linear Models Assumptions
Normally distributed outcome
Homogeneity of variances
Models are robust against this assumption
NOT required for 2 sample t-test if using unpooled variance
Violated?
Wilcoxon rank-sum test ~ Mann-Whitney U test ("t-test" )
Wilcoxon sign-rank test ("paired t-test")
Kruskall-Wallis test ("ANOVA")
Spearman rank correlation coefficient ("Pearson's correlation coefficient")
Binary or Categorical (propotions)
Independent
Risk difference/ Relative risk (2x2 table)
Z-distribution
Risk ratio
Odds ratio from logistic regression
Chi-square test (RxC table)
P(A)*P(B)= P(A&B) !Independence
Expected cell count =P(A)
P(B)
N
Chi-square distribution
Logistic regression (multivariate regression technique)
Correlated
McNemar's chi-square test (2x2 table)
Conditional logistic regression (multivariate regression technique)
GEE modeling (multivariate regression technique)
Alternatives if sparse data
McNemar's exact test ("McNemar's chi-square test")
Fisher's exact test ("Chi-square")
Time to event (Survial analysis)
Independent
Rate ratio (2 groups)
Kaplan-Meier statistics (2 or more groups)
Non-parametric estimate of the survival function
Empirical probability of surviving
Taking into account of censoring
Compare different group by log-rank test (a type of chi-square test)
Cox regression (multivariate regression technique)
Correlated
Frailty model (multivariate regression technique)