Importance of Effect Size Flashcards
what is effect size
objective and standardised measure of the magnitude of an effect
what is statistical significance? value?
P<.05
○ Estimates likelihood the relationship/difference/effect found is down to chance ○ Doesn't tell anything about size of relationship between variables
What does Benjamin et al., (2018) suggest about statistical significance?
○ Stricter alpha level of p<.005 as p<.05 results in a high rate of type 1 errors
○ (α): rejecting H0 when it was true (false positive)
○ Still arbitrary or should it be influenced by the goals of the researchers?
effect size and standard normal distribution
- Equivalent to z score of a standard normal distribution
○ Standardised: mean = 0, SD = 1
○ Describes a values relationship to the mean in terms of SDs from the mean- Mean of experimental group is located 0.2 SD above the control group
Cohens d effect sizes
small = 0.2
med = 0.5
large = 0.8
Pearson’s correlational coefficient: r
range
-1.0 to +1.0
cohens guidelines for r
small = 0.1
med = 0.3
large = 0.5
R2 – Magnitude of shared variance:
R2 is the amount of variance in variable B that can be attributed to variable A.
effect size guidelines for eta squared
small = 0.01
med = 0.059
large = 0.138
difference between eat squared and partial eta squared
partial eta squared is often much larger
they calculate the same for one-way anova
effect size guidelines for partial eat squared
small = 0.02
med = 0.13
large = 0.26
what are odds ratios?
Ratio of the odds of an event occurring in one group compared to another
odds ratio
OR < 1
Exposure associated with lower odds of outcome
○ As the predictor variable increases, the odds of the outcome decrease
odds ratio
OR > 1
Exposure associated with increased odds of outcome
As the predictor variable increases, so do the odds of the outcome
benefits of reporting effect sizes
- Resistant to sample size influence
- Encourages interpreting effects on a continuum
- Used to quantitatively compare results of studies completed in different settings
benefits of pooling effect sizes
help resolve inconsistencies in findings
pooling effect sizes (meta-analysis)
Estimating size of an effect in population by pooling effect sizes from different studies that test the same hypotheses
problems with cohens conventions for effect size
- Misleading labels?
○ One size fits all - Doesn’t consider potentially questionable research practices
Baguley (2009) suggests simple or unstandardised effect sizes may be a more suitable alternative
what is logistic regression?
- Regression with a categorical outcome variable/DV
○ Predictors/IVs can either be continuous or categorical
binary logistic regression - variables
2 dichotomous categories in DV (e.g. presence/absence of dementia)
multinomial logistic regression - variables
> 2 categories in DV
types of logistic regression
binary
multinomial
what does logistic regression tell us?
- Prediction of which, of 2 categories, a person is likely to belong to, given their scores on predictors
assumptions of binary LR
multicollinearity
linearity
independence
multicollinearity assumption of LR
○ Predictors/Ivs should not be too highly correlated with each other
○ Correlations above .7 are a problem
○ Check VIF and tolerance
§ VIF <10
§ Tolerance >0.1
§ These are ok
linearity assumption of LR
○ When outcome variable is categorical this is violated
○ Logistic regression assumes a linear relationship between continuous predictors and the logit of the DV
§ Must transform data using logarithmic transformation
§ Logit is the inverse of the standard logistic function
independence assumption of LR
○ Each observation must be independent
○ Observations should not come from repeated measures (before/after) or matched data
