Estimating Sample Size And Power Flashcards
What are the steps for estimating sample size?
- State the null hypothesis and a one- or two-sided alternative hypothesis
- Select the appropriate statistical test based on the type of predictor variable and outcome variable in those hypotheses
- Choose a reasonable effect size (and variability, if necessary).
- Set alpha and beta (specify a two-sided alpha unless the alternative hypothesis is clearly on-sided).
- Use the appropriate table or formula to estimate the sample size
Can the t test be used for a study with participants fewer than 30?
No
How can the effect size and variability be estimated?
They can be estimated by using previous studies in the literature and consultation with experts.
What is a standardized effect size?
It is the unitless quantity that makes it possible to estimate a sample size when an investigator cannot obtain information about the variability of the outcome variable. (effect size divided by the standard deviation of the outcome variable.
What is the shortcut formula to estimating sample size using a t test?
When more than 30 subjects will be needed:
Sample size=16/(standardized effect size)^2
What is relative risk?
It’s the risk ratio, is the ratio of the event of something occurring in an exposed group to the event of something occurring in a comparison, non-exposed.
Is correlation coefficient calculation important in clinical studies outside of behavioral medicine?
No
What does one get when we square the correlation coefficient?
Represents the proportion of the spread (variance) in an outcome variable that results from its linear association with a predictor variable, vice versa
True or false: Dropouts count in the sample size.
False
What is the equation to use when when accounting for lost follow-up participants?
1/[1-C]
C= estimated percentage of participants that will be lost
What is one thing that must be done to adjust for confounding variables?
Increase sample size
The magnitude of the increase in sample size when handling confounding variables depends on what factors?
- Prevalence of confounder
- the strength of the association between the predictor and the confounder
- the strength of the association between confounder and outcome
What is one statistical technique we can use to adjust for confounder and for differences in length of follow-up?
Cox proportional hazards
How many ordinal variables must there be to be treated as continuous variables?
six or more and if averaging the values of the variable makes sense
True or false: Some tests are more interested in finding out that the null hypothesis is true.
True
Equivalence studies have 1._______ power and 2.________ effect size which force it to have a 3._________ sample size.
- High
- Small
- Very large
What is an equivalence study?
A study in which the study should show no substantial association between the predictor and the outcome variable; the null hypothesis is correct
What is the main issue with equivalence studies?
The loss of usual safeguards. There is no Type I error.
What do descriptive studies lack?
They do not have predictor or outcome variables, nor do they compare different groups.
What do descriptive studies commonly report?
Confidence intervals. Investigator must specify desired level and width of the confidence interval.
If the sample size is fixed or limited, what must the investigator do?
Researcher must work backwards from the fixed sample size. Use sample size equations or work backwards using table.
What are the five strategies for minimizing sample size and maximizing power?
- Use continuous variables
- Use paired measurements
- Use more precise variables
- Use unequal group sizes
- Use a more common outcome
What is an unpaired two-sample t test?
A test that looks at the mean of two groups and assesses the difference between the groups
What is a paired two-sample t test?
It is a test that looks at the change in a pair of measurements in two groups separately
What is the difference between one- and two-sample t test?
One-sample t test only assesses one group. Two-sample assesses two groups.
When the outcome is dichotomous, using a ________ outcome, up to a frequency of .5, is usually one of the best ways to increase power.
More frequent
What are ways to make an outcome more common?
- Enroll subjects that are at greater risk of developing outcome
- Extend follow-up period
- Loosen definition of what constitutes an outcome
True or false: Studies with rare outcomes require large sample sizes.
True
When are pilot studies advisable?
Studies that involve new instruments, measurement methods, or recruitment strategies.
What trick can be applied to find out the standard deviation of a group of continuous variables?
S.D. Can be estimated as one-quarter of the difference between high and low ends of the range of values that occur commonly, avoiding outliers.
When the mean and standard deviation for continuous or categorical variable are in doubt one can ___________ that variable.
Dichotomize.
By doing so, this strategy avoids having to calculate the S.D. but still we must report proportions of the variable
Should the sample size be estimated early or late in a project?
Early when fundamental changes can still be made
Does the sample size account for those that need to be enrolled?
No. It accounts for number of subjects with outcome data.