EMB: Hypothesis Testing Flashcards
RR:
(population 1 (i.e. diseased) / total) / (population 2 / total)
RRI:
1 - (RR)
Absolute risk increase:
(population 2/total) - (population 1/total)
Number Needed to Harm
1 / (absolute risk increase)
Independent Variable:
Dependent Variable
I: Exposure (predictor, explanatory, risk factor)
D: Outcome (response)
Important Study Parameters:
- Types of intervention
- What outcome is being assessed
- What is the study population
Null Hypothesis
No difference between groups. RR=1
Alternative Hypothesis
Exposure is associated with disease.
Can be 1 or 2 sided:
RR>1 or RR doesn’t equal 1
Parametric methods
Require some assumptions
Compare different means
T-tests and ANOVA
Nonparametric methods
no assumptinos
categorical data
Chi-square test
compares PROPORTION between two (or more) groups. Categorical data (yes/no).
(male vs. female)
t-test
compares MEAN values between TWO groups
Continuous variable (all different age),
ANOVA
analyzes data that includes MULTIPLE variables
compare MEAN values
z-score
measure of standard deviation where data is transformed to standard normal distribution
Hypothesis test
compare test statistic to critical value (-1.96 to 1.96) if above or below, statistically significant.
This correlates with a P-Value
P-value
probabilty that an association as strong or stronger might have arisen by chance alone (IF NULL HYPOTHESIS WERE TRUE)
NOT probability THAT null hypothesis were true.
What must you consider when assessing Sample Size
- Type of Study
- Magnitude of difference (what will be clinically important by setting alpha and beta)
- *As difference desired is LARGER (very prevalent), you want smaller population
- *As ALPHA and POWER is smaller the sample becomes larger
- Random Error
- Power
Type I Error and Alpha value
Reject Null Hypothesis, but shouldn’t have
Alpha: probability of making a type I error (.05), smaller risk of being wrong because don’t want to change something if you’re wrong.
Type II Error
Beta value
Don’t reject Null Hypothesis, but should have. Often due to inadequate power (small population).
Beta: probability of making a type II error. (0.20)
Power
Ability of a study to detect a true difference between groups (1 -beta)
b= .2 then power =.8 detecting a difference
Point Estimate
- indicator of magnitude of effect
- must be supplemented with measure of random error AKA confidence interval
Confidence interval
Measure of Precision. 95% of the time this effect will happen. Lower Bound (Null Hypothesis1) ( your Point estimate) Higher Bound
Intention-to-Treat
“analyze what you randomize.” Include ALL participants in analysis, regardless of follow-up (increases generalizability)
Stratified Analysis
Examine exposure-outcome relationship in subgroups (male/female)
Adjusted Analysis
Control for effect of extraneous variables on the exposure-outcome relationship (control for confounding)
Confounding
Estimate of the exposure effect is distorted because it is mixed with effect of extraneous factor