BIOL 210 Lecture Notes - Lecture 18: Ronald Fisher, Cognitive Training, Meta-Analysis
11/09/2017
Lecture 18
• Correlational Studies
o Our six variables in Wennerson and Wold are dependent, because they depend on
other factors
o Asks if 8 other variables are potentially compounding
o A good correlational study can determine if a predictor has a causal effect, or if
variation is due to something else
o This study employs a multiple regression, where we can put in many variables
and examine a difference
o This study shows that the 8 other factors (nationality, letter of recommendation,
etc) are not confounding
• 5% threshold for p-value is arbitrary
o Made by some dude named Sir Ronald Fisher
o Symbol for >.05= NS
o However, just as .04 can be a false positive, .06 can be a false negative! One
might right “slightly above the threshold for statistical significance”
o Larger sample might be easier to show effect due to chance
• Friedberg paper: Evaeluation of Conflict of Interest in Economic Analyses of New Drugs
Used in Oncology (bad paper)
o Took a lit review of drug papers, separated papers as either publicly funded (n=
24) or privately funded (n=20)
o Two lead authors read all the papers, giving a conclusion that’s either favorable,
neutral, or damaging to the drug
o Two people could only agree on the classification 87% of the time, and gave the
third author the decision. However, the third author knew about the paper’s source
before they made the decision. Is this shady? Yes.
o The final p value was 0.4. Is THIS shady? (no, but it’s close to the border, let’s
take a closer look)
o This paper is bad. Why? Poor sample size, poorly blinded reviewers
• Four steps of a lit review:
o Identification (search database and reference lists)
o Screening (duplicates removed, records screened and excluded)
o Eligibility (Full-text articles assessed for eligibility: full-text articles excluded due
to non-human probands, insufficiently reported data, and irrelevant study design)
o Inclusion: Left with a smaller number that works
• Type 1 versus Type 2 errors
• Brain Size and IQ Study (good study)
o Excellent study in that it used a rigorous, systematic literature review & meta
analysis
o Finds p values that are far from the 0.05 threshold for false positive
o This study suggest that there’s a tiny effect, the tiny effect is likely overestimated,
and that it’s not due
o Thorough because it’s based on lots of studies, based on a huge meta-analysis
• George Mason Brain Study: Does study recruitment impact participant performance?
o Recruited in one of two ways: 1. For “Brain training” or for 2. “Study Credits”
find more resources at oneclass.com
find more resources at oneclass.com