Present: Levi Bonnell, Justine Dee, Nancy Gell, Kairn Kelley, Ben
Littenberg, Gail Rose, Connie van Eeghen
Start Up: NYT Article on When the Revolution Came for Amy Cuddy (2017) and
the reality of using P values
1.
P-Curve: A Key
to the File Drawer, Simonsohn, 2014
a. A P value has 3
parts: n, variance, effect size
b. If effect is 0,
then p will vary from about 0 to almost 1, evenly distributed, across all
possible values of p
c. If effect is
strong, then p will vary in a right skewed curve across all possible values of
p
d. Remember: the p
value is not about the hypothesis, it is about the data underlying the
hypothesis. It tells you if the data are
convincing but it does not tell you if the hypothesis is convincing.
2.
What are the assumptions that allow us to use a p value
a. The assumptions
of the test selected for analysis were met
b. Representative
selection of subjects
c. Independent
results
d. Focused on the
hypothesis
e. Methods were
conducted with integrity
f.
Asks one question, and one question only
i.
Alternatively, avoid the garden of forking paths to get one
through the garden to a gate you like
ii.
i.e. do not explore the data before testing the data
3.
Will Bayesian statistics fix this?
a. Not
necessarily. It may help, in getting rid
of the confidence interval process. It does not test how good the hypothesis
is.
b. What it does:
given what I know before the study, and what I learned from the study, here’s
the next estimate of the interval around the correct answer to the hypothesis
about the effect of interest.
a.
Nov 22: cancelled
b.
Nov 29:
Kairn Kelley – application for Pathways Mentorship Program
c.
Dec 6: Field
trip to Research Tapas on “Research and Reproducibility”
d.
Future
topics:
a.
Juvena:
protocol development
b.
LaMantia:
predictors of successful R01 applications: http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0155060
Posts
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.