related to RACE

Sex v. Gender v. Sexuality

However, race may be informative about the social experience a person may have within a mixed society

What are the biological and cultural influences on these perceptions?

Sexual dimorphism, pregnancy, breastfeeding

Assertive mating (culture based) supports inbreeding within mixed societies/cultures

We breed with partners that we have things in common with, including religion, physical traits, ages, SES, intelligence, politics, ect-

Making groups more alike over time

Example: sensitive caregiving v controlling caregiving

a moderate view in cultural research that maintains that behavioral phenomenon may be based on invariant psychological processes but that each culture will induce different manifestations of those underlying processes

Content Validity

Diagnostic interview Schedule

For western, middle class individuals this can be about generalized anxiety

For developing societies or individuals without running water and indoor plumbing it could be about resources

(Measure variables carefully)

Inferred relationships are often erroneous

-Random events do not look random

-Use formal statistical methods to test hypotheses about relationships

Relationships that do exist may not be causal relationships (Some research designs do not demonstrate causality)

-Different experiences & individuals

*Life sciences are more complicated than strictly physical sciences

-Studies may have different results due to sample size

*Although both report same p-values, one may be more significant (esp. with a bigger pop.) because it has more power & faith

-Relationship problems may be due to methodological issues

-A precedes B

-A and B must covary (B must occur when A does)

-A must be the most plausible cause for B with other potential causes ruled out

((The methods we discuss today do not meet these conditions. [correlational methods]))

When A changes, there must be concurrent or follow-up changes in B

A rational explanation with other potential causes ruled out [other potential causes is a criterion but have a theoretically, reasonable explanation is good but not always critical]

-(Gender – impossible to manipulate

Love/hate, alcoholism – unethical to manipulate)

Some processes take a long time to study experimentally (Language development in children)

Non-experimental methods can be used as a means of suggesting, clarifying, refining, or extending experimental research findings.

A correlational analysis quantifies the strength of the relationship between two or more variables

Where a correlation exists, it can be used to predict values for one variable from values for other variables (regression analysis)

______________________

Purpose: establish if a relationship exists and to describe that relationship

Measurements can be made in natural settings or in the lab, but we are simply measuring (at least) two variables to see if there is a consistent pattern of relationship.

-Assume: relationship b/wn variables

-Goal: Measure the strength of the association; how correlated are A &B?

-e.g. For the sample (not the individual) that if you know what “Mom’s level of depression is” then you can make a reasonable assumption about the “level of anxiety” their child has can’t say that anxiety

*Predict one group mean from the next

-Applied to samples, NOT individuals

Using a personality test to predict how someone will respond to certain situations

Using an aptitude test to determine whether applicants have the skills necessary for a job.

Gathering 2 or more bits of information from a person

No attempt to manipulate anything, just measured variables.

Self-esteem and attitudes towards minority groups

Class attendance predicting final grade

Statistics:

Can be used on any set of data, regardless of method

Can be used in a true-experiment, quasi-experiment, or a correlational study

____

-Correlations between a measurement and a measurement of itself again—reliability

No attempt to manipulate anything, just measured variables.

Self-esteem and attitudes towards minority groups

Class attendance predicting final grade

Can be used in a true-experiment, quasi-experiment, or a correlational study

A descriptive statistic that describes the linear relationship between two variables.

-Linear relationship: point A to point B that makes sense (a LINE)

Correlations range from -1.00 to +1.00

Absolute size indicates strength of the relationship

Sign indicates direction of the relationship

Scatterplots provide a pictorial view of the relationship

-Weakest correlation: (asymptotic) to 0

Is the p value less then the alpha level?

Magnitude and Sign of r

General guidelines by Cohen (1988) – use with caution

Small: |r| = .20-.29

Medium: |r| = .30-.49

Large: |r| = .50-1.0

P-value? Probability of achieveing a corelation this large or larger if the correlation in the population was zero.

If the probability is low, it means that there is little chance that the pop correlation is zero.

So we say that the correlation is statistically significant. (usually less than .05)

-Statistic tells you the likelihood that something is different than 0 it is ZERO

*Even if r = .8 but p = .06, the correlation is indeed 0

~Due to be underpower e.g. sample size = 3

-Small effects (can’t say certain relationships aren’t meaningful)

-Don’t know that something that is a small effect size will translate into something meaningful down the line

*Use with caution

Proportion of variance shared by the two variables you have correlated

Often more useful than the correlation

Note: r2 is always less than r in absolute value (except at extremes: 0 & +/-1)

-Covariation is r-squared (intersection of 2 variables in regards to a Venn-diagram)

*Accounts for that amount of variation in that particular relationship

. R-square

Common misconception that correlations are on a ratio scale. For example, a correlation of .8 is not twice as strong as .4. Or we cannot predict one variable from another twice as accurately.

Need to think more in terms of Variance.

When all the scores fall close to a line, then the variance is small (there is little error variance) and the correlation is high.

Squaring the correlation gives us a better measure of variance. SO if r=.8 (or -.8) then the r-square is .64 which means that one variable accounts for 64% of the variance in the other.

So in the example I gave you earlier .4 vs .8 (.8 gives you 64%) what does .4 give you? (16%) Not half at all.

A corr of .99 indicates 98% of the variability in the criterion can be accounted for or predicted by the predictor

Range Restriction (Truncated Range)

Heterogeneous subsets (multiple populations)

Outliers or Extreme scores

ALWAYS check to see if your relationship is linear before computing a correlation coefficient or regression analysis.

-Relationship BUT the LINEAR relationship here is 0

*Nonlinear relationship (power function—parabolic)

*Negative quadratic

-At any extreme in real life, things are never often good happy medium

For the full range of scores, r = .45

but for data in the restricted range (shown within the circle), the value of the correlation drops, r = .29

*Driven by age

-Within each grade, as anxiety goes then reading scores go down

Overall positive relation between two variables, but each of the subsets has a negative correlation.

When you combine subgroups into one overall data set, the correlation coefficient may not provide useful information.

-Be mindful of range (not only that there is an adequate range to capture all the variability & look to see the distribution e.g. floating point then possible censor/delete it)

*Treat extreme score as a case study –> pop up w/in samples as an extreme(s) then follow-up

Predicting values for one variable if we know the value of other variables

The predictor is the known variable. The criterion is the predicted variable

Predicting values for one variable if we know the value of other variables

The predictor is the known variable. The criterion is the predicted variable

A regression equation is an equation of the form

Y = a + b X

where X is the predictor, Y the criterion

*y = mx + b

Example of how correlational design can be used

Prediction

For example: relationship between certain behaviors and immenent suicide attempts (can use warning signs, like talking more about suicide or giving away prized possessions) and take steps to intervene

Doesn’t necessarily have to be about future behavior, for example,

Knowing someone’s IQ allows us to make some predictions about the intelligence of their parents. Use available knowledge of one variable to predict value of unavailable variable.

Often describe one variable as the predictor and one as the criterion. In cases of prediction the designation is quite clear. GRE to predict graduate school success. GRE predictor, success is criterion.

Y = a + b X

where X is the predictor, Y the criterion

The equation predicts Y if we know X

A common example: predicting GPA based on SAT score.

_____

Predicting scores on one variable from scores on another can be done mathematically through a regression analysis.

Use standard equation for a straight line.

Y is the score we want to predict (vertical axis),

A is the value of where the extended regression line would hit the vertical axis (the y-intercept),

B is the slope of the line

X is the score we already know

Use standard equation for a straight line.

Y is the score we want to predict (vertical axis),

A is the value of where the extended regression line would hit the vertical axis (the y-intercept),

B is the slope of the line

X is the score we already know

Beta is equivalent to the correlation

Example: predict college GPA from SAT scores and high school grades

Regression analysis can be extended to Multiple Regression

Using several predictors for one criterion

Extension is multiple regression in which we predict Y from a combination of two or more predictors

This cannot only be more accurate (rel’ps are complicated) but it can also

Tell us whether using 3 predictor variables accounts for more variance in the criterion variable than just 2 or one.

ALSO

It can be used to statistically control for potential third variables.

This is the correlation between X and Y, with Z held constant

Partial correlations are easy to calculate

Mediation/Moderation (Baron and Kenny, 1986)

Mediator: a variable that accounts for the relationship between 2 other variables.

Moderator: Specifies when effects will hold

Moderator: Specifies when effects will hold

-Controls

-Partial: controlling for another variable & seeing if it might mediate

*Form of multiple regression

In one study, r = 0.45

A little thought suggests that there is a critical mediating variable: Age

Age and language skills correlate 0.65

Age and big toe size correlate 0.62

The partial correlation of language skills and big toe size can be calculated to be 0.08

___________

When the correlation drops it is a “mediating variable”

When it turns out to be larger, then it is referred to as a “suppressor variable”

-Age is accounting for the shard covariation

__

When the correlation drops it is a “mediating variable”

When it turns out to be larger, then it is referred to as a “suppressor variable”

Small, significant effect size is more important no matter how large r is

When the correlation drops it is a “mediating variable”

When it turns out to be larger, then it is referred to as a “suppressor variable”

Survey question= Item

It is standard to provide anonymity and confidentiality to respondents

LECTURE

Confidential- I know who you are, but no one else does

Anony- researcher doesn’t even know who you are (cant use with longitudinal research- cant follow up!)

Ironically, stressing anonymity and confidentiality may arouse suspicions in respondents

Incentives (v coercion)

Make compliance the path of least resistance

Use the foot in the door phenomenon (small request, get them in a request- ask you a couple question it will be really quick…now I have just a few more)

Use the door in the face phenomenon (give them something that they are obvi going to say no to and will say yes to in comparison)

LECTURE: cant guarantee that youll get the number or demographics that you need to answer your question

tendency for people to agree to a small request after turning down a larger request

Ex: Social Desirability concerns

How do we fix the problem?

Make sure that confidentiality and anonymity is made salient

1. People don’t always believe this and sometimes people don’t want to admit things to themselves

2. Ask or Prime for honest responses &

Generate rapport

“we know that many people find it difficult to use a condom every time they have sex. What we would like to know is how many times this has happened to you in the last year.”

3. Use a deception check

Ex: “have you ever told a lie?”

We may not know the reason for our own attitudes and behaviors

OR, we may simply not know the answer

People are often embarrassed about not knowing the answer.

EX. Questions about politics

How do you feel about the agricultural trade act of 2001?

30% of a sample typically provide an answer to issues that are invented by the researcher

Solution?

Can be reduced by an explicit “don’t know” alternative

EX: Gallup uses “no opinion/don’t know” response

Avoid double-barreled questions

Avoid loaded/leading questions

Avoid negative wording/difficult to understand

Avoid acquiescence (yea, nay saying)

Avoid vague response options

Beware of order effects (response options/questions)

dont switch scale ranges, avoid range restrictions

Ask sensitive questions sensitively

Use questions relevant to all

Striking the right balance for number of questions/items

Order of Questions: the answer to a question might be influenced by the question asked before it

Fixing order Effects

Counter balancing- changing the order person to person (random order, or say half gets this half first and blah blah to see if there IS an order effect); no way to get rid of it within a single person, but you don’t survey just one person

Order effects should even out

Can also check to see if there is an order effect in your data analysis

Most of data will be clustered at top/bottom (avoidable), now you don’t have range/variablilty

Number that makes you think metrically that you may be uncomfortable with (sometimes build in artifact, like putting a letter grade to it)

Improve phrasing to improve responses

More items= more reliability

The better the overall response

How many is too many?

Pretest and try it out yourself and with pretest subjects

Up front, so that you have the answers if they stop filling out the survey?

If they are boring (demographic) questions, will they stop before getting to the “good” stuff?

Group items together logically

According to topic

According to response format (rating scale, open-ended)

Consider the lay-out (simple and uncluttered)

Vocabulary and language should be appropriate

Example slide 26: what to eat on a date

________

Observational, archival and survey research can be very useful and provide valuable information

__________

Often a good starting point for developing hypotheses

Appropriate caution must be used because of the limitations of these methods.

Caution: sometimes a third variable can account for the association between two variables because the third variable is directly predicting both of the others

Mediating v Confounding

Con: Expensive, difficult

can prevent from making valid effects across groups

____

Particularly problematic when you have only 1 sample/1 division

Def: Differences between age groups based on when they were born

When comparing with age based differences

Raw cultural changes, some slow, some fast- affect individuals differently & generationally over time

9/11, PTSD:

1 singular even changes group dramatically…

o Ex: VOTER TURN OUT

More 18-24 year olds voted in the 2008 election than previous elections (2004-previous)

Not the same group of ppl, even if in same window of people

Not interested in following same group of people, not interested in individual change

Interested in group (population-level) change

o Basically cross sectional study that you repeat over and over again,

o Relatively cheap

o Committed to following particular group over time

o NOT new sample each time in field (compared to trend study)

o Concerns: decide who they are (bc once you recruit, youre stuck with them),keeping track, ect

o Common to select based on age (group of two year olds, levels of aggression, ect)

o Can still have cohort effects (sample of 2nd grade teachers- could all be different ages, others have been teaching for years, ect)

o Way to try yo get around cohort effects by creating multiple samples

o Follow each sample, but examine

Implications of discrete event

o Military- define cohorts by war, for instance- can have veterans all the same age that are have been in four different wars

Ex) sexual abuse, ptsd, schizophrenia

Problem: memory is fallible , accuracy of the reports are questionable

Things that are low-base rate, like schizophrenia/bipolar- they are the more extreme conditions

Con: COST, cohort effects, 1-shot esp when it costs 1B

Sample size, what can you get, when can you get, money, concern of validity of measurement

Retrospective report is easier

Most at risk to drop: the risky, interesting people- boring ppl stick around

Results in decreased power

Relates to variation and sample size

When you lose the extreme groups, you lose variation & everyone starts to look alike

Over sample the group you think you’re going to lose

Common mistake- representative sample at time 1, but at time 4 it’s no longer representative- should have over sampled extremes so at time 4, it is representative- bc not as interested in where they start, interested in where they end up FRONTLOAD

Diff stats for cross sectional & longitudinal

Key c

Independent t-test: samples must be independent from one another & randomly assigned

LONgitudinal data analysis- sample is NOT independent of one another (at diff times, still same group of people) cross-sectional-ANOVAs

Lack of independence (more challenging), not true of trend studies (bc you don’t have same ppl, independent groups)

Multiple imputation- statistical processes, half my data is great, half is missing,

Can estimate/derive MY missing values given other data to figure out what I should have gotten based on what I have done already

Purposely do not collect data on certain percent of the data bc we can impute the missing values, so instead of seeing 1000 ppl at each assessment point, I see 800 people (20%less)

Choice made in beginning of study