Research and stats Flashcards
1
Q
2 types of extraneous variables
A
1) confounding - related to IV
2) disturbance - DV
2
Q
- percentages of scores in each standard deviation in normal distribution
A
1 = 68%
2 = 95%
3 = 99%
3
Q
Leptokurtic vs. Platykurtic
A
Lepto = sharp peak “piled up” in the middle; flatter tails than normal
Platykurtic = flatter in the middle; thicker tails
plat rhymes with flat
4
Q
History threats
Maturation threats
Differential selection
- how to control
A
- 1+ groups
- random assignment to groups
- differential selection = random assignment
5
Q
Statistical regression aka regression to the mean
- how to control
A
1+ groups and ensuring their extreme scores are equivalent at the beginning
6
Q
THREATS
- testing-
- instrumentation
- differential attrition
A
- pretest affects post test (don’t administer pretest or use Solomon group)
- instrument used to measure dv changes over time
- drop out reasons
7
Q
Types of validity
Population
Ecological
Temporal
Treatment variation
Outcome
A
- different population
- environment
- across time
- variations in treatment
- different but related dvs
8
Q
Reactivity
- demand characteristics
- experiment expectancy
A
- act different
- cues from research
- experiment behaviour is biased
9
Q
- multiple treatment interference aka carryover or order effects
A
- counterbalancing
10
Q
- selection treatment interaction
A
- people differ from the population and the difference affects how they respond to the IV
11
Q
Pretest treatment interaction
A
- participants respond to the iv differently
12
Q
- Solomon 4 group design
A
IV= expose 2 groups to IV and only give one the pretest
DV= don’t expose groups to DV and one group takes and pre and post test and other group takes the post test only
13
Q
Grounded theory
Phenomenology
Ethnography
Thematic
A
- abstract theory via participants views
- lived experience of ppl
- join culture
- find themes
14
Q
Single subject designs
(Can be 2+ subjects but then treated as a single group)
- AB
- reversal
- Multiple baseline
A
DV measured multiple times
- if change is due to maturation
- ABA, ABABA (withdraw after AB) if change is due to history
- IV applied to multiple different baselines
15
Q
Group designs (ANOVA)
- between subjects
- within subjects
- mixed designs
A
- 2+ groups exposed to 1IV with 2+ levels (what therapy affects anxiety cbt, eft, or psychodynamic)
- within subjects experiences all levels of IV
- include tome series design (measure DV multiple times before IV)
- 2IV (I between and 1 within) example - CBT, eft, psychodynamic and time (pre, mid and long ost)
16
Q
Factorial designs
A
- 2+ IV with 2+ levels each
- include main and interaction effects
Eg. Therapy type and gender
17
Q
Probability sample
- simple random
- systematic random
- stratified random
- cluster random
A
- everyone has equal change
- select every th
- divide everyone into subgroups based on characterisitcs then random select
- natural clusters or random sampling not possible - select a sample of clusters or random sample of individuals
18
Q
Non probability sample
- convenience
- voluntary response
- purposive/judgemental
- snowball
A
- ready accessible
- volunteered
- research used judgment to select
- referral based
19
Q
Bivariate correlations
- Pearson
- spearman
- point biserial
- biserial
- contingency
A
- both are continuous and linear relationship (use eta when not linear)
- both ranks spear killed moby dick which is a high ranking book
- 1 continuous; 1 true dichotomy he’s the true one point blank
- 1 continuous; 1 fake dichotomy
- both nominal
20
Q
Multivariate correlations
- multiple regression
- canonical correlation
- discriminate function
- logistic regression.
A
- 2+ predictors, continous
- 2+ predictors, 2+ criteria, all continous
- 2+ predictors, 1 nominal criteria discriminate = you need to discriminate the criteria
- discrimination function for non normal data
21
Q
SEM
- observed
- latent
- exogenous
- endogenous
A
- directly observed and measured
- inferred
- not explain or predicted in sem
- explained by other variables in sem
exogenous = EXclude
22
Q
Null vs alternative hypothesis
A
- iv doesn’t have affect (you want to retain a true null hypothesis)
- iv does have an effect (you want to reject a false null hypothesis)
*false null hypothesis means the null hypothesis is something else (alternative hypothesis = there is an effect)
23
Q
Type 1 error
Type 2 error
A
- reject a true null hypothesis (false positive) - alpha (signifier)
- retain a false null hypothesis (false negative) - beta
24
Q
What is Bayesian statistics
A
- combine info from data with previous info/beliefs/certainty about parameter
- use % chance and % confidence intervals
25
26
Bayes theorem
3
1) previous knowledge (prior)
2) current data (likelihood function)
3) update knowledge (posterior)
27
Parametric vs non parametric
- interval and ratio data vs nominal and ordinal data
28
- Single sample chi square (goodness of fit)
- multiple sample (contingency tables)
- 1 variable (if girls like lipstick or lip gloss)
2+ variables (if girls and woman like lipstick or lipgloss)
29
Single sample t test
T test for unrelated/uncorrelated samples
T test for related/correlated samples
- compare sample to population mean
- 2 unrelated groups
- 2 related eg. Matched participants, pre-post
30
ANOVA
- 3+ levels of an IV or continuous DV
- f ratio > 1 = effect
- mean square between / mean square within
31
Other ANOVAS
Factorial
Mixed/split plot
Randomized block
ANCOVA
MANOVA
Trend analysis
- 2+ IV (better than separate f ratios for each main and interaction effect)
- 1 between subjects IV and 1 within subjects IV
- controls for variable by including it as an IV
- controls for variable by removing its effect on DV
- 1+ IV and 2+DV (continuous variables)
- if IV has significant linear or nonlinear relationships to DV
32
F ratio
Planned comparisons
Post hoc tests
- significant main effect
- based on theory
- after f ratio is significant
33
Cohens d
- small, medium, large
- less than .2
- .2 to .8
- more than .8
34
Clinical significant (Jacobson Truax)
- calculate a relianble change index (RCI) to see if significant is due to actual change rather than error
- calculate cut off
- recovered, improved, unchanged, deteriortaitsd
*is JW actually going to change??**
35
Reliability
- test retest
- alternate forms
- internal consistency
- inter rater
- scores over time
- scores across different forms
- scores over test items
- across raters
36
- internal consistency
- coefficient alpha/cronbachs
- kinder Richardson’s
- split half reliability
- spearman’s brown prophecy formula
- average inter item consistency across sample
- dichotomous items
- split the test in half and correlating the 2 scores
- determined effects of lengthening or shortening test items
*you’re crude and you’re rich. You’re either rich or not*
37
Inter rater
- percent agreement
- cohens kappa
- % agreed (doesn’t account for chance agreement)
- nominal *I used this for my qualitative analysis*
38
Factors affecting reliability
- content homogeneity
- range of scores
- guessing
- homogenous = larger
- less restricted range = larger
- easier to guess
39
Items difficulty
Item discrimination
- % answered correctly (0-1)
- ability to discriminate high and low performance (-1 to +1)
40
Confidence intervals
- calculate standard error of measurement
- SD X square root of (1-reliability coefficient)
- 68% confidence interval = 1 standard error
- 95% = 2 standard error
- 99% = 3 standard errors
41
Item response theory
- item characteristic curve (x and y axis)
- difficulty/secerirt/location parameter
- discrimination parameter
- probability of guessing correctly
- relationship between item and latent trait
- x axis = total test scores
- y axis = probably of answering correctly or endorsing items
- level of trait required for 50% to endorse items
- discriminate between high or low of the trait; steep = better
- 0 = more difficulty
42
43
Define cross validation and shrinkage
- when is shrinkage greatest
- related to criterion validity
-cross validate using a new sample because shrinkage can happen (some correlations happen by chance and these likely won’t happen again during the second sampling)
- greatest shrinkage when sample is small and there are 2+ predictors
44
Standard error of estimate
Confidence interval around predicted criterion score
45
Correction for attenuation
- criterion related coefficient gets low because of measurement error
- so validity coefficient underestimate relationship
- correction estimates the maximum validity coefficient
46
Define these for predictor and criterion/incremental validity
- true positives
- false positives
- true negatives
- false negatives
- base rate (how to calculate)
- positive hit rate (how to calculate)
- high P and C
- high P and low C
- low P and C
- low P and high C
- base rate (# true positive + false negatives/n)
- positive hit (# true positives/total # of positives)
47
- define and formula
- sensitivity
- specificity
- hit rate
- positive predictive value
- negative predictive value
- people with dx (true positive/true positive + false negative)
- people without dx (true negative/true negative + false positive)
- correctly categorized (true positive + true negative/n)
- probably test positive actually had dx (true positive/true positive + false positive)
- probably test negative actually not have dx (true negative/true negative + false negative)
48
99-11
- 95 CI = 2 SD above and below mean
- standard error of 3
- 3x2 = 6
49
Relationship between reliability and validity
As validity increases, reliability shrinks
