Potential Errors In Epidemiologic Studies II.
Random Error
Dr. Sherine Shawky
• Understand the concept of random error • Recognize the methods to prevent random error • Know the methods to evaluate the role of chance on results
Learning Objectives
Performance Objectives
• Improve precision • Evaluate the role of chance
In most epidemiologic studies, it is impossible to evaluate every member of the entire population. Thus, the relationship between exposure and health-related event is judged from observations on sample of the population
Samples
n1 n2 n3
n4
n6 n5
N
Chance Random Error Lack of Precision
Control of Random Error
Prevent Study
Evaluate
Prevention of Random Error
Hypothesis
Sample size
Type of Error
Hypothesis
H0 = No difference H1 = Some difference
Types of Error
Study results Do not reject H0 Reject H0 H0 in reality True False Confidence level (1- ) Type I error () Type II error () Power (1-)
Sample Size
n1 n2
N
How many subjects are required ?
Sample Size Calculation
Assumption Parameters Factors
Assumption for Sample Size Calculation
H0 is not true & H1 is true
Factors for Sample Size Calculation
• Population size • Research question • Study design • Type of data
Parameters for Sample Size Calculation
• Probability of type I error • Probability of type II error • Proportion of population that are exposed to, or have healthrelated event • Magnitude of the expected effect
?
What is the power of this study if only these subjects are available ?
Power
Power Calculation
Work the appropriate sample size equation in the inverse direction, using the available sample size
Evaluation of the Role of Chance
Statistical Testing Confidence Interval
Statistical Testing
Assumption P-value Statistical test
Assumption for Statistical Testing
H0 is true
Choice of Statistical Test
• Research question • Type of data • Characteristics of data
P-value
• The P-value is the estimated value for issue from results • The P-value depends on the sample size and the strength of the association
P-value (cont.)
• Two-tailed for given magnitude and uncertain direction • One tailed for given magnitude and known direction
Confidence Interval (CI)
• More informative than P-value • Indicates presence or absence of statistical significance • Calculated for mean, proportion, relative risk and odds ratio
Interpretation of CI
Not significant Mean/ Proportion (one sample) Mean/ Proportion (two samples) Relative risk/odds ratio Significant Value is Value is not included in CI included in CI Two CIs overlap Two CIs don’t overlap
1.0 is included 1.0 is not in CI included in CI
Conclusion
When a research is performed on a sample of the population, the researcher has to minimize the role of chance before initiating the study. Also, he should evaluate its impact on the results before making decisions.