statistical power Flashcards
whats the most commonly used inferential statistical method
hypothesis test
whats does the hypothesis test begin with
a null hypothesis
whats the purpose of the hypothesis test
to rule out chance as an explanation of the results
what does the outcome of a hypothesis test depend on
the sample size
what happens if you increase the sample size of a hypothesis test
increases the likelihood of obtaining a significant result
what is recommended that researchers provide when reporting a statistically significant effect
an independent measure of effect size
type 1 error
rejecting the true hypothesis
type 11 error
accepting a false hypothesis
what is statistical power
the probability of correctly identifying an effect
-rejecting a false null hypothesis
when would researchers attend to power
at the design stage
what can power analysis be used for
to calculate the minimum sample size required to accept the outcome of a statistical test with a particular level of confidence
practical significance
a result or treatment that is large enough to have value in practical application
what is needed for practical significance
the size of the effect
what is a statistically significant result
one thats unlikely to be due to chance
what is a practically significant result
one thats meaningful in the real world
effect size
the measured magnitude of a treatment effect or relationship that is not influenced by factors such as sample size
what is the estimation of effect size essential for
- practical significance
- desired sample size
- comparison across different studies
the d family
effect sizes assessing the difference between groups on continuous variables
what are effect sizes for continuous variables
standardised mean differences
how is effect size for continuous variables calculated
subtract mean of one group from the other and divide result by the SD of population
*the bigger the score the bigger the effect
cohens d
the size of the mean difference between two treatments can be standardised by measuring the mean difference in terms of SD
what does d mean in cohens d
sample mean difference divided by sample SD
what does d=2 in cohens d indicate
mean difference is twice as big as SD
what does d=0.50 indicate in cohens d
mean difference is half as large as the SD
the r family
effect sizes measuring the strength of a relationship between 2 or more variables
what are egs of the r family
- correlation coefficients
- standardized regression weights
- regression measures of variance explained
context of interpretation of effect sizes
small effects can be important if they trigger big consequences, accumulate into larger effects or lead to technological breakthroughs and new discoveries
contribution to knowledge - interpreting effect sizes
- interpret results in context of current evidence
* does the observed effect differ from what others have found
whats a small standardised mean difference for cohens criteria
.20
whats a medium standardised mean difference for cohens criteria
.50
whats a large standardised mean difference for cohens criteria
.80
whats a very large standardised mean difference for cohens criteria
1.30
whats a small correlation for cohens criteria
.10
whats a medium correlation for cohens criteria
.30
whats a large correlation for cohens criteria
.50
whats a very large correlation for cohens criteria
.70
how to increase statistical power - effect size
increasing effect size increases power
how to increase statistical power - alpha level
- usual value p= 0.05
* increasing alpha increases power
how to increase statistical power - sample size
increasing sample size increases power
what are the 4 main parameters of power analysis
- effect size
- sample size
- alpha significance criterion
- power of the statistical test
how to obtain the effect size for power analysis - a literature review
- estimate the effect size based on published studies similar
- relevant values are not entirely clear
how to obtain the effect size for power analysis - a pilot study
- rough estimate of effect size
* info extracted from a small sample is limited and bias
how to obtain the effect size for power analysis - cohens reccomendations
*small effects - large effects reccomendations