Lecture 3 Flashcards
What would the distribution of p-values drawn from repeated samples from the same population look like if the null or alternaltive hypothesis is true?
When the alternative hypothesis is true, more p values are closer towards 0 (skewed).
When null hypothesis is true, the distibution is flate, but there is some unavoidable Type 1 error (where p < 0.05)
Define Dependent variable (DV)
The variable that you are trying to understand and predict; The psychological measure you collected
For basic models what conditions must the DV meet?
The DV must be in a continuous scale and there must be only one DV
Define Independent variable (IV)
One or more variables used to improve the accuracy of your prediction of the dependent variable
GLM:
DV = B0 * IV1+ B1 * IV2 + Noise
B = paramter
What is dummy variable coding? What does it imply
dummy variable coding - manually defining IV = 0
if subject i is in Group 1 and IV = 1 if i is Group 2, then DV = β0 + IV · β1 + noise. This implies:
* Group 1 has a population mean β0
* Group 2 has a population mean β0 + β1
* H0 : β1 = 0 is equivalent to a two-sample t test.
(One-/two-sample, paired) t test, correlation, linear regression, and ANOVA can be summarized into:
Fitting a GLM + Applying a F test
One-sample t test - It fits a horizontal line and tests for the intercept
Correlation and linear regression can be used to find the slope
When to use what effect size?
If you are comparing mean differences between two groups A and B, use Cohen’s d
If you conduct correlation or simple linear regression, use Pearson correlation ( r )
If you conduct one-way ANOVA, and assuming that your sample size per group is the same and K is the number of groups, use Cohen’s f
Effect size
Effect size is a standardized numeric constant that characterizes the signal-to-noise ratio (SNR) of your interest.; effect size is a value measuring the strength of the relationship between two variables in a population, or a sample-based estimate of that quantity
* Approximate of how well your sample statistics captures the population parameters
How can you use effect size to compute power?
Using effect size to compute power. You can obtain an (approximate) effect size using your data. Combined with the sample size used in your study, you can also get an approximate power
In other words, “if someone replicates my study, what is the probability that they will get the same statistical result?”
G Power needs: Type 1 error threshold (decision rule), effect size, your study design, and sample size
How can you use effect size to compute sample size?
Using effect size to compute sample size. If you have specified your study design and obtained an appropriate effect size, you can compute the minimum number of samples to achieve certain amount of power
G Power needs: Type 1 error threshold (decision rule), effect size, your study design, and wanted power
Again, the effect size is computed from the data, and does not reflect the population effect size precisely. In the Effect size simulation slide, your reported effect size could be as large as 2.7 by chance although it is 1.0 in truth. In such a case, your power is exaggerated that what it should be. Try to plan off the “Smallest Effect Size Of Interest” (SESOI)
Why report effect size?
It will imply readers of (i) the expected power of your study and (ii) scientific importance
