Week 4 Flashcards
According to Simmons et al. (2011), what is the most costly error?
a false positive, the incorrect rejection of a null hypothesis
What does it mean to set the significance value at p ≤ .05?
the maximum false-positive rate is 5%
how do DoF make false positives more likely?
researchers don’t preregister designs, so they can explore various analytic alternatives to search for a combination that yields statistical significance and then report only what worked
which are the 4 most common ways to choose your DoF?
- choosing number of dependent variables
- choosing sample size
- using covariates
- reporting subsets of experimental conditions
Why is flexibility in sample size a problem?
because there is no linear relationship between sample size and likelihood of significant results
What 6 requirements do Simmons et al. (2011) propose for authors?
- decide rule for terminating data collection beforehand & report it
- at least 20 observations per cell
- list all measured variables
- report all experimental conditions including failed manipulations
- if observations are excluded, report results for when they are included also
- If an analysis includes a covariate, also report analysis results without the covariate
What are nonsolutions for the problem of researcher DoF?
- correcting alpha levels
- using bayesian statistics
- conceptual replications
- posting materials and data
Why doesn’t correcting alpha levels help the DoF problem?
- it is unclear which and how many DoF contribute to a given finding and what their effect is on the false-positive rate
- there is no rule about how to adjust each DoF
Why don’t Bayesian statistics solve the DoF problem?
it increases researcher DoF:
1. constitutes a new set of analyses that can be applied flexibly on data
2. require adjustments on a case-to-case basis
What solution do Wagenmakers et al. (2012) propose for minimising false positives?
preregistration
what is exploratory research?
hypotheses are formed after data analysis
what is confirmatory research?
the entire analysis plan is made explicit before any analysis is done
what is a methodologically sound way to combine exploratory and confirmatory research?
collect the data all at once and then split the data into exploratory and confirmatory subsets
What are Latin squares?
partial counterbalancing method for condition order - each condition appears once/participant and once/position in the sequence of conditions
What is a limitation of Latin squares?
possible sequential effects (condition B always follows condition A)