Class Notes (1,100,000)
CA (630,000)
UTSC (30,000)
Psychology (8,000)
PSYB04H3 (100)
Lecture 12

PSYB04H3 Lecture Notes - Lecture 12: Diederik Stapel, Mmr Vaccine Controversy, Null Hypothesis


Department
Psychology
Course Code
PSYB04H3
Professor
Connie Boudens
Lecture
12

This preview shows pages 1-3. to view the full 11 pages of the document.
Beta, Alpha, Power Clarified
o As alpha increase, power increases
Type I error: concluding there is an effect when there is none
o Incorrect rejection of the null
Null hypothesis assumes there's no effect
Require high level of proof to reject null hypothesis and go with alt. hypothesis
Alpha (α) = probability of Type I error (blue area below)
Probability of incorrectly rejecting null hypothesis
o Researcher sets alpha (.05 or .01 are common)
This is set in advance
Leads to:
o (1 α) = probability of correct decision when the null is true (i.e. no effect exists in reality)
o Null hypothesis is probably true, results not unusual enough to occur by chance and has no
effect
Type II error: concluding there is no effect when there is one
o Incorrect failure to reject the null (null is true)
Incorrectly accepting null hypothesis (?)
Beta (β): probability of a Type II error (pink area below)
Leads to:
o (1 –β ) = probability of correct decision when the null is not true (i.e. there is an effect in
real life)
o Probability you'll find it
(1 –β ) is power
o Find evidence for a difference when it exists in the real world
o Something a researcher wants to maximize
Studies have low power because not enough participants or looking for a small effect
Below
o Pink part is probability of type 1 error
o Bell curve on left is what would be if null hypothesis was true
o Dotted line is cut off line (represented by alpha)
Belongs to curve on the right
o Pink area is beta (to the left)
o Power is everything to the right of dotted line in the second curve
o If move to the left, making alpha bigger
Bugger alpha is and because you're willing to accept a risk
Accepting a lower bar/standard
Going to wind up with decisions that aren't different then null hypothesis
Will conclude alt hypothesis has more support (probably going to be wrong)
o But if increase alpha
Pink are to the right will be smaller and white are will be bigger
The more power under the alt hypothesis, the between
o As alpha get bigger, power gets bigger too
find more resources at oneclass.com
find more resources at oneclass.com

Only pages 1-3 are available for preview. Some parts have been intentionally blurred.

What Makes a Study Important? - Replicability, Generalizability, Realism
Important to know
Ex. If it's in the news and published in high quality journal, that doesn't mean that its an important
research finding
There's still questions that you need to ask 1430
When Alpha Is Not Firm
Alpha conventionally set at .05 or .01
o p < .05 or .01 for effect to be considered statistically significant (value that you need)
Certain practices change that probability
o Even if researcher decided on .05 0r .01, when a studies being done. Stuff happens that
changes the level
Decisions researchers make, such as: (based on article)
o Control variables?
Extra things that were measured
Such as gender, age, income etc.
Problem is if they include these measures in their analysis (not actually measuring
them)
o Comparisons?
Ex. 3 groups
What will they compare? 1 and 2, 2 and3?
o Analyses?
Which variables they're going to correlate with each other
Which variables to put in their program
o Dependent variables?
Put a bunch in cause don't know
Problem is reporting only on the statistically significant dependent variables
and not all of them
o Subsets of the variables?
Independent variables and dependent variables
o Outliers?
It's a data point that's unusual and far from the rest of the pattern
Researcher has to decide what to do with it
It's a judgment call
find more resources at oneclass.com
find more resources at oneclass.com

Only pages 1-3 are available for preview. Some parts have been intentionally blurred.

In theory, supposed to report what they do with the outliers, but it may not actually
happen
o Its okay if these decisions are made in advance
Not all these decisions are made in advance
o If they add additional analysis, will increase alpha beyond what they set
Measuring many variables, only reporting results for some relationships
o Ex: measuring heart rate, skin conductance, self report for anxiety
Lets say looking at public transit and anxiety: look at relationship between the two
things
Found that heart rate and conductance goes up but not anxiety
Person may choose not to report the finding
Reporting data from some groups, not others
o Ex: groups 1 and 3, leaving out group 2
Only reporting difference with group 1 and 3 as if person planned to do all along
Problem when researcher changes it. Its okay if that's hat they decided before hand.
Collecting data, analyzing results, adding participants and re-running analyses
A-posteriori hypotheses
o Hypothesis after the fact
o Supposed to be set in advance
o Maybe researcher accidently ran analysis and found something: cant add that into ongoing
hypothesis
o Have to set hypothesis before hand
Fishing in the data
o Looking into the data before hand
o Higher chance of significant effect (for all these things: when check again and not update it,
say then that that's your hypothesis or answer)
The good news:
o Calls for raw data to be shared
Post it where people can look at it and do their own analysis
o People more aware of retraction
When someone finds a mistake, journal is supposed to retract article and print a
statement
Happened with relationship between autism and vaccines
There's actually no relationship
Website on research articles that have been retracted
Fraud
Reputable people look at everything, such as overall alpha rate vs just one test
o Ex: Diederik Stapel fabricated data for at least 55 publications while at 3 universities
Does a brain train
find more resources at oneclass.com
find more resources at oneclass.com
You're Reading a Preview

Unlock to view full version