The flashcards below were created by user
Rx2013
on FreezingBlue Flashcards.

power is a calculation that
indicates whether or not a study can accurately detect statistically significant differences between samples when a difference exists

simply stated, power =
1  beta

want power to be at least
80%

a priori calculation of power
number of subjects needed to meet power is figured out before the study is performed

post hoc calculation of power
 number of subjects needed to meet power is calculated after the study has been performed
 potential for bias
 could be manipulation just to get published

If a statistically significant difference between groups is detected
power is less of an issue b/c power is the chance that we will find a difference

study did not enter enough pt to meet power but difference was detected?
info we can use

study had enough participants to meet power and showed a difference between groups?
info we can use

study had enough pt to meet power and showed no difference between groups
info we canuse

study did not have enough to meet power and results did not show difference between groups
 may experience type II error
 study did not meet power

power is mentioned but without details, no % set or number of pt required to meet power
power not met

power is claimed to be met but not reported
power not met

Three types of analysis
 per protocol
 intent to treat
 modified intent to treat

per protocol
number of subjects completing the trial

intent to treat (ITT)
number of subjects randomized into each group

modified intent to treat
number of subjects randomized and met prespecified criteria (ex. 8 weeks of tx)

descriptive statistics are calculated to describe characteristics of
 a group
 ex. patient demographics & frequency

measures of central tendency

mean
 used on interval or ratio data
 may be misleading on ordinal data
 affected by outliers

Median
 used on ordinal
 nonparametric test

mode
 variable that occurs most frequently
 most useful with nominal data

measures of variability
 useful in measuring how close data is to the measure of central tendancy
 range
 interquartile range
 variance
 standard deviation
 standard deviation of the mean

Range
 degree of spread
 influenced by outliers
 difference between largest and smallest observed

interquartile range
 used for ordinal, interval and ratio data
 range of values remaining when the largest and smallest 25% are removed

Variance
 gives more info of the data set's variability
 measures avg squared distance from the mean
 often more useful to use standard deviation to express variability

Standard deviation
 square root of variance
 interval and ratio data
 only useful with normally distributed data

Standard error of the mean
 derived from SD
 SEM is always smaller than SD
 greater N = smaller SEM
 used to calculate confidence intervals

Confidence interval
 indication of the outcome within the population
 range of values in which the true value is included
 ex. 95% sure that the range of values contains the true value
 descriptive
 help interpret clinical significance of data

the width of CI measures
 reliability of sample data
 wide interval = less reliable
 small interval = more reliable

if the CI crosses zero with ordinal and interval data
 possibility that there was not difference between treatment groups
 should not be reported as statistically significant

if the CI crosses 1with ratio data
possibility that there is no difference between treatment groups

advantages of CI
 may expose manipulation
 help in determining clinical significance
 compliments Pvalue

if trial shows there is no statistical significance between two groups, pay attention to
upper end of CI

if trial shows that there is a statistical difference between two groups, pay attention to
the lower end of CI

