psyc 012

Observational research

systematically recording behavior of peoples and animals. it happens in public, laboratories, and classrooms. it is done on paper and pencil, audio recording, or video recording. works on frequency, association, and causal. works because it avoids problems with self-reports(rationalization, social desirability, inaccurate memories, and participants who cant speak(babies and animals)).

Naturalistic observation

observe behavior as it naturally occurs; no or minimal intrusion(animals in the wild or online social network data)

Participant observation

researcher becomes part of the world they're observing

Contrived observation

observe behavior in research settings(lab observation)

Undisguised observation

participants know they're being observed. ex: kinder kids in class

Reactivity

people may change their behavior when they know they're being observed (parents talking more when a camera is present vs. audio recording only) (also affects construct validity)

Disguised observation

participants do not know they are being observed.

Partial concealment strategy

participants know they're being observed, but not why

Observer bias

observers' expectations influence their interpretation of behavior. ex: gender and aggression... children in snow suits and if participants were told to male then they counted aggressive acts and if female it wasn't. (affects construct validity)

Observer Expectancy effects

observers' expectations actually influence participants behavior. ex: Rosenthal and Fode's "maze bright" vs. "maze dull" rats. all rats were the same but those with the bright rats "performed" better due to bias. (affects construct validity)

Masked/Blind bias

observers unaware of conditions and/or study purpose. Prevents observer bias and observer expectancy effects but not reactivity.

Solution for reactivity

blend in, record(with permission), test over long periods(people forget), and avoid video/experimenter pressure.

Codebook

list of definitions for scoring behavior; improves Inter-rater reliability. ex: defining exactly what counts as looking at face vs hair. but high inter-rater doesn't mean bias free since two can have the same bias.

threats and solutions

(threat)Observer bias --> (changes)experimenter --> (cause) experimenter expectations --> (solution) masked/blind design.

observer effects --> participant --> experimenter expectations --> masked/blind design

reactivity --> participant --> being observed --> covert observation.

population

the entire group of interest

census

measuring every member of the population

sample

a subset of the population

generalize

to apply findings from sample to the population

unbiased/representative sample

all members of the population have an equal chance of being included

biased sample

some members have a greater chance of being included

matters

bad sampling is a threat to external validity

sampling most important

most important in frequency claims.

not important in association claims or causal claims.

convenience sampling

using people who are readily available(college undergraduates and Mechanical Turk(problem is that its not representative follows WEIRD.. also MTurk not good cause unemployed and tech savvy people. (biased)

Self-selection

sample only contains people who choose to participate (tend to be people with strong opinions and have more free time) (biased)

Snowball sampling

participants recruit other participants(social networking to get participants) ex: a friend tells a friend and then so on (biased)

Purposive sampling

only recruiting particular types of people in a non-random way. ex: posting ads at baby store to find pregnant women. (biased)

quota sampling

fill categories(like stratified) but using a non-random way of recruitment. ex: having a sample of 13-17yrs old and going to the mall and asking the first 20; 13 or 15 yr old you see and disregard the other 13 or 15 if quota is filled. (biased)

simple random sampling

assign a number to everyone, use a random number generator (unbiased/representative)

Cluster sampling

randomly select clusters then test every person in clusters. ex: randomly pick 5 schools out of 10 and then in the 5 schools you test every single person in that school

Stratified random sampling

randomly select individuals within different categories(strata) to make match population distribution. ex: you have 100 ppl and 60% of the population is magic and the other isn't. you take a sample of 10 and in your population you get 6 magic people and 4 non magic.

systematic sampling

every nth person. ex: you have 110 people and you need 10 groups.. then each group has 11 people and then you randomly select the 7th person and then add 11 and you get the 18th person and so on.

multi stage sampling

randomly select clusters and then randomly select a subset. ex: school example as well but randomly select clusters in that school.. clusters are interchangeable subsets used for for convenience. strata are meaningful divisions where you must sample from all of the to be representative.

critical of sampling concepts

larger samples are not more representative. 100 truly random people are better than 1,000,000 non-random people.

sample size affects: statistical validity(margin of error(+/- percentage).

sample-selection(subset of individuals from a larger population)affects: external validity(generalizability)

experiment

manipulating at least 1 variable(IV) and measuring at least 1 variable(DV). not everything in a lab is an experiment. (ex: measuring steps at two ages, testing video game players and non-video game players.) if nothing is manipulated it's not an experiment.

independent variable (IV)

what was manipulated

dependent variable (DV)

what was measured

Control variable

variable that is held constant.. ex: beanbag weight.

control group

Receives a placebo, standard treatment, or no intervention at all. (between subjects)

experimental group

Receives the treatment, intervention, or specific condition you are testing. (between subjects)

conditions/levels

different groups of the IV. ex: light bat, normal bat, and heavy bat.

IV and DV

IV comes first since its the cause and DV comes after cause its the effect. IVs are on the x-axis and the DVs are on the y-axis.

comparison group

another level of IV and its also an alternative treatment

covariance

first step required to prove that X causes Y... groups differ.

levels of the IV provide a comparison group(two groups one is IV other is comparison and you see if there is a change); DV must vary to IV(switch from comparison to experimental(IV) the DV must change as a result).

temporal precedence

IV occurs before before DV is measured. IV is first.

internal validity

no other variable (confounds) can explain changes in DV. selection ruins internal validity in between groups(random assignment).

design confounds

an experimenter makes a mistake designing an IV so on second variable varies systematically with the IV. ex: Team A uses a normal bat and Team B uses a heavy bat and faces a better pitcher. can ruin internal validity if 2nd variable varies systematically with IV.

selection effect

participants in one level of the IV are systematically different from participants in another level. ex: Group A is champions while Group B are last. solution would be random assignment(equal chance to each level)

Matched Groups Procedure

measure a potential confound then assign participants to create groups that evenly distribute that variable. (ranking batters and distributing top batters evenly across groups)

systematic

confound varies along with the IV (bad)

unsystematic

confound is distributed across the IV levels (ok).

lumping

putting all of one thing into a one level of the IV. ex: all coffee drinkers happen to be morning people and the water drinkers(comparison group) happen to be night owls. you wouldn't know if test score was coffee(IV) or morning people(confound).

spreading

opposite of lumping. ex: morning and night people spread evenly through both groups IV and comparison group. So confound cancels out.

Posttest-only design

randomly assign to IV condition --> receive manipulation associated with their IV--> Measure DV (between subjects) ex: randomly assign, one group gets IV level 1 and the other level 2, after that the DV is recorded. ex: good Samaritan study

pretest/posttest design

randomly assign --> measure DV(pre) --> manipulation --> measure DV(post) (between subjects) pros: rules out pre-test differences, allows observation of change. cons: more steps, might contaminate via repeated testing. randomly assign then measure the DV(pre) and the manipulate the IV and then measure the DV (post). ex: GRE study

within groups or within subjects

each participant is exposed to all levels of IV. only one group of participants because each participant completes every condition.

repeated-measures design

one group is exposed to a sequence of IV manipulations. ex: heavy bat --> size estimate --> normal bat --> size estimate. Condition A then condition B.

concurrent measures design

participants exposed to all levels of the IV at the same time. Condition A and Condition B at the same time. ex: infants looking at female vs. male faces side by side.

Advantages of within groups/subjects

avoids selection effects(participants are their own control group), increases power(ability of a study to show significant effect(reduces unsystematic variability)), more efficient(fewer participants needed).

Disadvantage of within groups/subjects

order effects which is exposure to one condition changes how participants respond to another(fatigue, practice, carry over). Demand characteristics which is participants might detect the study's purpose and change behavior. solution is counter balancing(randomly assigning participants to experience conditions in different orders) counter balancing doesnt fix order.

practice effect

long sequence might led to participants to get better at task

fatigue effects

people get tired or bored towards the end

carryover effects

some form of contamination carries from one condition to the next

Types of counterbalancing

full counterbalancing(all possible orders are presented)

partial counterbalancing(only some orders are presented)

Latin square design(a type of counterbalancing in which each condition appears in each position once).. Full: 1,2,3. partial: 2,3,1. Latin: 3,1,2.

Manipulation check

measure another DV to quantify how well the IV worked. ex: survey participants after.

Pilot Study

conduct manipulation check with a separate group.

external validity in experiments

generalizing to other people(random sampling) and other situations. Random assignment is about internal validity, not external validity. researchers often prioritize internal validity over external validity in experiments.

effect size

strength of the difference between conditions

Cohen's D

standardized effect size(mean difference/standard deviation) Larger d= larger effect. ex: d= .2 is small, d= .5 medium, d= .8 large.

statistical significance

how likely is our result due to random chance

null hypothesis

the hypothesis that there is no significant difference between specified populations, any observed difference being due to sampling or experimental error.

reject the null means yes significance.

fail to reject the null means no difference between groups

p-value

probability of getting the result if the null is true. Convention: p< .05 is significant... p> .05 not significant

t-test

used for a simple experiment(2 conditions).. large t-value requirements: large mean difference, small variability, large sample size.

Null effects

1) not enough difference between groups: weak manipulations, insensitive measurement, ceiling effects(scores squeezed at high end), floor effects(scores squeezed at low end).

2) too much within-group/subjects variability: measurement error, individual differences.

3) sample size too small: individual differences skew results.

4) there really is no effect.

power: the probability of finding significant effect when the effect is "true" > 80%

Six Threats to Internal Validity over-time (pretest/posttest)

1) Maturation: gradual change overtime(not due to treatment). ex: clinical trials w/o a placebo or control; some people will just recover from illness regardless treatment.

2) History: external event affects most members at the same time at the same time as treatment. ex: life disruption from moving home interfered with exam 1.. better adjust before exam 2, better results.

3) Regression to mean: if a score is at pretest extreme, the posttest score is likely to be less extreme.

4) Attrition: participants drop out(systematic attrition is bad; unsystematic is ok)

5) Testing threat: change as a result if taking a test more than once(practice/fatigue/carryover) ex: WALDO

6) Instrumentation Threat: The process of measuring the DV changes over time(coders criteria)

coding criteria

The specific rules or guidelines a researcher uses to decide what counts as a specific behavior and what doesn't. is in the codebook and protects inter-rater reliability and construct validity.

bad criteria --> low inter-rater reliability --> low construct validity.

good criteria --> high inter-rater reliability --> high construct validity

Threats to any design

Observer Bias: researchers' expectations influence interpretation.

Demand characteristics: participants change behavior because they know what the study is.

Placebo Effect: Knowledge of receiving treatment leads to outcome. solution: double-blind study(neither observes or participants know the conditions)(need control to see placebo effect).

Pretest: symptoms before manipulation. 3 IV(no therapy, placebo, therapy). Posttest symptoms after therapy.

Factorial designs

can experiment with two or more independent variables(factors)

condition

one level of an independent variable.

Crossed design

every combination of the IVs are tested

Cell

a particular combination of the conditions of each IV

2x2 design

2 IVs, each with 2 levels= 4 conditions made.

Main effects

the overall effect of one IV on the DV, averaging across all levels of the other IV ex: a study of coffee, water, and ppl who ran or sat. you look at the coffee variable and the water one see the main effect of drink on heart rate. The running and walking one is you measure heart rate and see which had the main effect.

Interaction

when the effect of one IV depends on the level of the other IV. Differences in difference.

*the two variables combine to create a unique result that you wouldn't expect by looking at either variable entirely on its own.* ex: alcohol and sleeping pills

Moderators

is an interaction affect is essentially a test of moderation(A causes B, but only under certain conditions)

Simple effect

a comparison between two cells.

Types of interactions

Crossover(the effect of one IV reverses across levels of the other iV. one line has a positive slope, one has a negative slope(x)) and Spreading(the effect of one IV is stronger at one level of the other IV than the other level. one line has a slope, the other is either flat or in the same direction but flatter. no interaction, line parallel)

ANOVA(analysis of variance)

tests each main effect and the interaction for statistical significance(f statistic w/p-value). (t-tests of many variables and if there is any difference that exists among your groups). *statistical validity in factorial designs*

F-statistic

ratio that compares the variation of between groups to different groups to the variation of within groups to those same groups.

marginal means

used to determine main effects(averaging across IV)... simply the average score for one variable, completely ignoring the other variable. *statistical validity in factorial designs*

difference scores

used to determine interactions (comparing the difference between cells) *measure change over time*

nuisance variables

If order interacts with the main IV; counterbalancing failed, and the order of the conditions is now a confound.

Types of factorial designs

Independent-groups: each cell has a different group of participants.

Within-groups: one group completes all combinations of IV levels.

Mixed design: one factor is between-subjects, one is within-subjects.

2x3 design: 2 IVs, 2 levels(one variable has 1 and another has 2) and 6 conditions.

Advantages of Factorial Design and Testing Interactions

Interactions test theories: multiple variables relate; not describing how each variable affects an outcome, test how variables interact when determining.

Interactions test generalization of results: whether the effect of an IV on a DV is the same for different groups of participants. Mixing measure and manipulated IVs(experimental IV is manipulated... participant group is measured). Interaction says how effect varies for different populations(external validity)

Quasi experiments

researchers identify the IV but do not manipulate and do not randomly assign participants to conditions. why quasi.. when you can't ethically or practically manipulate the IV(age, sex, medical conditions).

non-equivalent group design: compares two groups that differ on a key variable(between/independent subjects).

non-equivalent groups pretest & posttest: measure the groups before they differ(pretest) then compare them after they differ(posttest).

Internal validity: always lower in quasi experiments than in true experiments.

selection effects are a major problem. solutions: matched group design and waitlist design.

systematic attrition is a problem because confounded with conditions.

matched groups design

equate participants on a key variable(verbal IQ in studies of autism vs. typically-developing kids)

waitlist design

all participants experience the treatment/manipulation, one group has to wait(randomly wait).

N-designs

each participant is treated as an experiment; individual data is reported rather than group averages.

single N-design. 1 person= case study(Patient HM)

patient HM

He lost his ability to form explicit memories (LTM) after his hippocampus was removed, but he could still form procedural memories (motor skills). This proved that memory is not just one system.

Stable-baseline design..(n-design)

observe a behavior for an extended period before treatment to establish a baseline. then test for change.

Reversal design...(n-design)

observe behavior with treatment, then take it away. If treatment is working, original behavior should return.