PSYCH 310 Unit 2

confounds

A variable that happens to vary systematically along with the independent variable

between subjects (aka independent groups)

All independent variable are between subjects

non-direction When we compute a correlation, other variables are treated equally Correlation is standardized You can easily compare 3 different correlations, even if the variables involved are different Correlation is restricted to 2 variables correlation is restricted to 2 variables works on ly on continuous variables (unless you do a point-biserial correlation

Correlation

It only equals 1/3 of causation, which is not enough. temporal precedence internal validity association.

Correlation does NOT equal Causation. Why?

moderator the graph doesn't explain why that relationship exists, so its not a mediator. It's not a third variable because if it was, when you include the variable, the relationship would go away. (still not sure bout the last part. )

Divorce among early married young adults, measuring religion. Is this mediator, moderator, or 3rd variable?

Err on the side of no. Longitudinal studies do no control for potential confounds (problem with internal validity)

Do longitudinal studies provide causation?

a 3rd variable that changes the relationship between two variables. (think debate moderator between trump and hillary-made sure they didn't kill each other)

Moderator

Moderators: gender of the follower.

Moderators: When do people hold the door open? Hypothesis: people hold the door open more often when the person behind is closer. Procedure: walk around campus following people through doors either closely (3 ft vs 10ft)

Researcher's expectations influence their interpretation of results.

Observer Bias

people receive a treatment and really improve- but only because they believe they are reviewing a valid treatment.

Placebo effect

dependent variable measure only once.

Post-test only design

Dependent variable measures twice (before and after)

Pretest/posttest designs

They may have been randomly matched (You accidentally got all 4 boys in one group and all 4 girls in the other)

Random assignment does not always work as well with small sample sizes. Why?

subgroups that cause two things to look like there is a correlation, when it is actually caused by something else (my definition) Ex. reading correlated with height. The actual cause is age. 6 year old are short and can't read as well. 10 year olds are taller and can read better.

Spurious corrlations

Systematic: a potential confound is correlated with the independent variables. (very very very bad because it destroys internal validity) Unsystematic: differences in some variable exist but are unrelated to the independent variable. (not a problem for internal validity. May impair statistical validity).

Systematic Vs. Unsystematic variability

covariance (A is associated with B) Temporal precedence (A comes before B) internal validity (C, D, E, are controlled)

The 3 things required for causality

Because there really is no effect! The different levels of the independent variable weren't enough There was too much unsystematic variability.

Wh night you get a null effect?

gender or parental marital status culture or gender gender

What are some potential moderators for? Effect of violent video games on later criminal behavior Effect of makeup on perceived likability? Effect of literary fiction on empathy?

Control:Not playing soccer Comparison: breastfeeding vs other feeding

What is the control or comparison in these scenarios? Playing soccer leads to increased use of metric system Breastfeeding may boost children's IQ

How strong was the result? How big was the difference?

What is the effect size?

3x2

What kind of factorial design? high school students, college, and college grads read wither literary fiction or non-fiction, and their empathy was measured?

1. At least 1 manipulated variable. (an independent variable) 2. At least 1 measured (dependent) variable. 3. Other variables are help constant (control variables). BE ABLE TO DISTINGUISH BETWEEN MANIPULATED AND MEASURED.

What makes an experiment an experiment?

simple regression

When two variables are correlated, we can use one to predict the other Ex. height at age 2 can be used to predict height age 18

multiple regression y=a + b1x2 + b2x2

When we can use more than on variable to predict another. To determine height at age 18, we may use height at age 2, foot length at age 2, parent's height.

multivariate design

When you measure each variable twice over time

construct and statistical The stats must be accurate. We don't need to establish cause and effect, so we don't need to eliminate other causes yet (so not internal validity)

Which are the two most important validities to interrogate for an association claim?

All of the above.

Which of the following can obscure the true different between groups? a ceiling effect on the dependent variable Unsystematic variability within each group Large measurement error All of the above

Manipulation was not strong enough Measure was not sensitive enough Ceiling or floor effects.

Why might the different between groups be too small?

to test limits: Factorial designs can explore whether an effect generalized across groups (external validity) Do children and adults both like clowns?

Why use factorial designs?

scatter plot correlation or regression

With association claims, you graph the data using a ________ ______ Statistically test the data using a ____________ or ___________.

mediator

a mechanism that explains how one variable influences another

a moderator if there is a moderator, then the association is true some cases (single moms) but not for others (married moms) moderators are relevant to external validity while subgroups are reliant to statistical validity.

a third variable that changes the relationship between two variables.

restriction of range

a variable does not vary enough; some possible values for the variable are left out.

r=.1 small or weak r=.3 medium or moderate r=.5 large or strong

correlation effect size conventions:

autocorrelations (auto-self)

correlation of a variable with itself across time

cross-lag correlations

correlation of one variable at time1 with the other variable at time 2

coree-sectional correlations

correlations of the two different variables measured at the same time points

unlikely to happen by chance the probability that your sample's association came from a population where the association is 0.

definition for statistically significance

participants guess what the study is supposed to be about and change their behavior in the expected directions.

demand characteristics

same as double blind, but the control group receives a placebo, such as a sugar pill, fake therapy, etc.

double blind placebo-control study

neither the experimenters nor participants know who is in what groups.

double-blind study

differences are different.

interaction

are the averages different? Are the young drivers different than the old drivers? You pretend the other variable doesn't exist.

main effect

longitudinal design This allow us to establish temporal precedence, because we measure the variables at different time points

measure the same variables in the same people at different points in time.

no significant different between the levels of the independent variable

null effect

confounds

occurs when you think one thing cause an outcome but in fact other things changed, too, so you are confused about what the cause really was.

mixed design

one variable or more is between subjects, one variable or more is within subjects

What is the relationship between effect size and statistical significance?

p-value

within subjects

participants rated likability and trustworthiness of attractive and unattractive faces that were wearing different amount of makeup (no makeup, a little, moderate, a lot)

internal validity sampling deals with the external construction.

random sampling is to external validity as random assignment is to _________ ________

regression is directional One variables is the predictor and there other is the dependent variables non-standardized (unless all the variables have been standardized, you ann't compare different regression coefficients) con income multiple predictor variables (but only one dependent variable) Regression can include both categorical and continuous variables as predictors

regression

curvilinear

A relationship might not be linear, so using statistics that look for linear relationships would be inappropriate.

main effect

A statistically significant effect of each independent variable separately, ignoring the other independent variable

everyone is maxed out not he measure Everyone gets close to 100% correct on all questions, regardless or reading difficulty.

Ceiling effects

each participant experiences only one level of the independent variable Each person is compared with someone else Random assignment is essential.

Characteristics of between subject designs

randomness!

How do we defend against systematic variability (confounds)?

As least two variables Both are measured-none are manipulated -it doesn't matter what kind of variable these are, or what kind of stats are used. If both variables are measure, then it is a correlational study making an association claim. (if a variable is manipulated, it is an experiment)

How do we know that a study is trying to make an association claim?

Measurement Error- your measure is not precise enough Individual Differences- people vary too much Situation noise- the environment was not carefully controlled.

How might unsystematic variability lead to null results?

temporal precedence: we don't know which came first. internal validity: We already have the association.

In the example of ice cream sales and shark attacks, what else is needed for this to be a causation.

"controlled for" "talking into account" "correction for" "after accounting for" "when [variable] is held constant"

Language of people talking about multiple regression results

Mediator: Variable 1 influences variable 2 BECAUSE of Mediator. The relationship is not direct. (Ex. lower income leads to higher divorce rates because making less money is stressful) (Ice cream sales lead to increased shark attacks because shows like ice cream) Moderator: Variable 1 influences variable 2 differently for differently levels of moderator. Relationship is direct. (Ex. Lower income is more strongly associated with divorce rates in rural that in urban communities). (ice cream sales lead to more shark attacks on children but do not influence the number of attacks on adults.) Third Variable: Variable 1 and 2 are really related to third variable, not to each other. Relationship isn't direct. (Ex. Lower income and higher divorce rates are both really related to education levels.) (increases in ice cream sales and shark attacks are really related to the temperature outside.

The difference between mediator, moderators and third variable confounds.

within subjects

all independent variables are within subjects