Class 1 Notes - The Power and Effect Size in Psychology

POWER & EFFECT SIZE RE-CAP P is the probability of observing results as observed, if null hypothesis is true. Inferential tests - How likely we get results if there is no effect (if NH is true) To unlikely we get a test statistic of magnitude w/ less than .05 Attribute the unlikely event to an effect in data - prob. of result occurring by chance is small Statistically Significant If Alpha level or .05 prob is large - it meant no effect on DV = Statistically non-Significant INFLUENCE OR NOT? Analyses & how big result was Inferential test is dependent on many other factors Two things can influence whether we find a significant difference b/w 2 means are the size of the effect & the size of sample chosen SIZE OF THE EFFECT EX. amount of time spent revising has huge effect on exam scores, difference b/w conditions is so large didn't occur by chance? so effect size is important SIZE OF SAMPLE Sample size interact w/ effect size - We decide that there has been a significant effect more likely to find that difference b/w means significant if sample is large EFFECT SIZE AND POWER NHST worn in conjunction w/ effect size Effect size Standard measure of magnitude magnitude of difference b/w conditions difference it makes to DV score estimate effect size from prev. research useful for estimating sample size avoiding Type II errors compare expected size across studies = help decide how many participants should recruit for any particular study risk of making either type I (finding real finding when we don't) or type II (have real finding, but we don't find) effect size stops type II error Power capacity to correctly reject null hypothesis sensitivity to relevant effect size ability to detect effect Ex. Bandura's theory that seeing aggressive role models = small effect on children's aggressive behavior Mean score of 13 & 17 based on small sample (3 in each group) = diff. is down to chance If big effect is there but sample (n) is small - explainable by other factors Large likelihood to find difference b/w mean that is sig. if sample size is large Large effect Less crucial sample size Big difference b/w means - less trivial LD ex: effect size is overtly large - less crucial sample size Significant sample size The bigger sample size, even if the effect is very small Significant findings can be found if: The sample is very big, even if the effect is very small Or effect is very big, even if the sample is very small EFFECT SIZE & STATISTICAL SIGNIFICANCE One variable has a statistically significant effect on another - how large the effect was Not called large, medium, small effect sizes Often times effect sizes measured in terms of d - effect size, hedges g, Pearson correlation coefficient r Cohen's d: small effect = 0.2, medium effect = 0.5, large effect = 0.7 When comparing 2 levels of an IV (line research) need to find difference b/w 2 means divided by standard deviation. Equation is used when investigating change b/w 1 central condition & 1 experimental condition No clear control group? To be pooled SD - n's number of ppl in group 1 & 2, sd's mean scores in groups To sum the variances of the groups & divide by 2. When square root Equation for calc pooled SD Effect size useful b/c objective measure of importance of effect LD Ex: study on the effect of interference on memory? reports found no effect - want to see if we need more participants If small effect size need more participants If IV has a small effect on DV, design study has a small sample size, diff. b/w conditions due to chance when they're real - a type 2 error POWER Power - probability of detecting the effect of IV on the DV, detecting effect size Avoiding type 2 errors Low power dangers of committing type 2 errors - unaware is there an effect? we failed to do so To appropriate sample size by using statistical power can call power analysis - after non-significant result Measured on a scale of 0-1 Low power probability ranges 0-1. Power of 0.1 chance of detecting an effect, power of 1 is detecting effect