Session 5 - Confidence Intervals Flashcards
(35 cards)
how is a 95% CI estimated?
by calculating 2 SE around the obtained value
What happens to the CI if the SD increases?
the CI increases.
What happens if the low range of a CI is a negative number?
then the finding can not be statistically significant because the null value lies in the CI. it’s like saying you are not 95% that the treatment works since a negative value indicates that it does NOT work.
why does the mean value or point estimate always fall within th 95% CI?
because the CI is constructed around the point estimate.
what does a broader CI reflect?
Less certainty or more potential error in your “best estimate”
more likely that the null will be in the interval
what does the curve width reflect?
the precision of the estimate
greater standard deviation or standard error=
greater variance among subjects.
more subjects=
greater approximation to the population=
more certainty of study value.
increased sample size=
decreased width of CI
what does a narrower CI reflect?
greater confidence in the point estimate.
a smaller effect size=
a greater likelihood that the null will be in the CI
Confidence Intervals are affected by
Variance
Sample size
Magnitude of effect (effect size)
what happens to the CI if you want to go from 95% to 99%?
the interval widens to increase the likelihood that the actual value lies within the range
what is the problem with having a wider CI?
it’s more likely that the null value will fall within the range and is therefore less likely to be statistically significant.
why would CIs be preferable to a p-value?
better visualization of possible fluctuation
doesn’t decide (dichotomize)like a p-value
gives you everything a p-value does plus more.
Power
how sure you can be that you will be able to rule out chance as an explanation for your findings.
Power=1-beta
do you estimate power before or after an experiment?
after
Do you estimate power when you have a null result?
yes.
you want to find out if it is a true null; false negative or true negative
what was the probability of a statistically significant result
power is a function of
(magnitude of effect)x alpha error (type-I risk)x sample size)/variance
what is a commonly accepted level of power?
80%
does 100% power mean that you will find the effect in your sample?
no.
100% power means that IF the effect you’re looking for is there, it’s likely to be statistically significant.
what is a limitation to absolute effect size?
it does not permit comparison of findings on a given intervention between 2 or more studies.
Standardized effect size
brings the absolute effect sizes of studies into a common set of units
only for parametric data where different outcome measurements were used.
how do you compare effect size for dichotomous data?
ARR - absolute risk reduction and RRR - relative risk reduction permits comparison of effects between studies with non-parametric outcome data. requires that data be dichotomized
