lecture 11- effect size and power

What are odds ratios?

• often used in clinical research

• How much does an intervention/risk factor change the odds of belonging to one group vs another?

• Interpretation: the risk factor increases the odds for the condition by a factor of X

• 1 = equivalent to no effect

• Small effect size: OR = 1.68

• Medium = 3.47

• Large = 6.71

Why are effect sizes useful?

• make results across studies comparable

• Meta-analysis- do effect sizes differ between studies? (Heterogeneity) Does effect size differ depending on certain study characteristics?

What is statistical power?

The probability of detecting an effect and obtaining a significant result if the alternative hypothesis is true.

• probability is higher is effect size is big

• Smaller in effect is small

How do you increase statistical power if we can’t change effects sizes?

larger samples- better able to approximate population value

• smaller standard errors

• Narrower confidence intervals

Higher significance levels.

What is adequate statistical power?

Cohen’s suggestion- 80%.

Reality- often much less.

To achieve adequate power, do a-priori power analysis to determine what sample size is needed.