Reading...
Play button
Play button
Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
115 Cards in this Set
- Front
- Back
|
Independent variable
|
"X"
Treatment/intervention must have 2+ levels to have comparison point |
|
|
Dependent Variable
|
"Y"
Observed outcome of treatment |
|
|
Protocoal Analysis
|
a type of content analysis
"heeded cognitions" that underlie problem-solving person asked to think aloud and their verbalizations are recorded verbalizations are categorized (intention, cognition, planning,evaluation) |
|
|
Interval recording
|
obsere a behavior for a period of equal-interval time periods
record if specific behavior occured during that period. Good for complex interactions and behaviors without concrete beginning and ending (laughing, playing, talking) |
|
|
Event Sampling
|
observe and record behaviors each time they occur
Good for behaviors that occur infrequently or have a long duration |
|
|
Situational Sampling
|
Observe behaviors in a number of settings
increases generalizability of findings |
|
|
Sequential Analysis
|
code behavior sequences (not just specific behaviors) to study complex social behaviors
|
|
|
True Experimental Research
|
random assignment to groups
enough control of variables to determine that a change in the DV is caused by the IV |
|
|
Random Assignment
|
randomly assigning participants to the various groups
|
|
|
Quasi-Experimental Research
|
experimenter does not have control over group assignment and must use preexisting groups
|
|
|
Simple Random Sampling
|
Every member of the population has an equal chance at being a part of the sample
Selection of one member of population has no effect on the liklihood of another members selection reduces probability that sample will be biased, especially if it's a large sample |
|
|
Stratified Random Sampling
|
Used when population varies in characteristics and to ensure each characteristic is represented in the sample, the population is split into "strata" and subjects are randomly selected from each strata
(i.e. gender, age, SES, race, culture) |
|
|
Cluster Sampling
|
Select units of individuals from population and either include all people in unit or randomly select from the unit
Useful when it is not possible to identify or gain access to entire population |
|
|
Causes of variability
|
1) independent variable (experimental variance - you want to maximize)
2)systematic error (you want to control) 3)random error (you want to minimize) |
|
|
extraneous variable
|
source of systematic error
-irrelevant to purpose of study, but confounds results because it correlates with the DV |
|
|
Random Assignment and variability
|
used to control equalize the effects of variability because all groups are the same on that extraneous variable
|
|
|
Ways to Control Variability from extraneous variables
|
-randomly assign to equalize
-hold extraneous variable constant by making all participants the same on this variable (limits generalizability) -match subjects on EV - matching is used to put in pairs and then randomly assign pairs, especially in small samples -Blocking: build variable into study as an extra IV, participants are "blocked" into groups based on status of EV and then randomly assigned -Statistically control it, especially in quasi-experiments, ANCOVA |
|
|
Blocking
|
when an extraneous variable is identified prior to starting, you can group participants into "blocks" based on the EV and then randomly assign to include the EV as another IV so that it can be analyzed.
|
|
|
Random Error
|
goal: minimize
-make sure participants dont get tired, experiment site is free from distractions and all measures are reliable |
|
|
Internal Validity
|
1) Is there a relationship between IV and DV?
2) Is the relationship causal? Threats to internal validity: maturation history testing instrumentation statistical regression selection attrition interaction with selection |
|
|
Maturation and internal validity
|
any biological change that occurs to participants during course of study that is not relevant for research
CONTROL: include more than one group & randomly assign, single-group time-series design so that DV is measured multiple times to gain information on maturation |
|
|
History and internal validity
|
when an external event effects DV
Most problematic if only have 1 group and the event occurs at the same time as IV. CONTROL: have more than one group |
|
|
Testing and internal validity
|
exposure to the test may effect performance when the test is readministered
CONTROL: administer DV measure only once, design a measure that minimizes memory and practice effects |
|
|
Instrumentation
|
changes in the accuracy or sensitivity of your device
CONTROL: more than one group, use same measures on all participants, ensure measuring devices do not change during study |
|
|
Statistical Regression and internal validity
|
extreme scores will likely regress toward the mean
CONTROL: don't include only extreme performers when selecting participants |
|
|
Selection and internal validity
|
If there are systematic differences between the groups because of selection process there will be low internal validity
CONTROL: administer pretest to determine initial similarity between groups, randomly assing |
|
|
Attrition and internal validity
|
threat to internal validity if drop-outs signficantly differ than those who remain in the study
CONTROL: use a pretest to be able to assess differences |
|
|
Interactions with Selection and validity
|
when groups are nonequivalant (i.e. one group exposed to historical event and the other was not)
|
|
|
External Validity
|
Can these findings be generalized?
*always limited by internal validity, however high internal validity doesn't guarantee external validity* Threats: interaction between testing and treatment interaction between selection and treatment reactivity/demand characteristics Multiple treatment interference |
|
|
population validity
|
can results be generalized to other people
|
|
|
Ecological Validity
|
can results be generalized to other settings
-especially important for lab studies |
|
|
Interaction between testing and treatment
|
threat to external validity
-administering a pre-test may "sensitize" subjects to the topic area adn therefore interfere with their reaction to IV limits generalizability to only those who were pretested CONTROL: don't administer pretest, SOLOMON 4 GROUP DESIGN (to assess impact of pretesting on internal and external validity by using pretest as another IV) |
|
|
Interaction between selection and treatment
|
threat to external validity
those who included in research responding to IV in a particular way based on their specific characteristics (i.e. those who volunteer are different than those who didn't -maybe more motivated) CONTROL: ensure sample is representative of population |
|
|
Reactivity
|
threat to external validity
-respond to IV in certain way because they know they are being watched (experimental conditions influence bx) EVALUATION APPREHENSION: behavior influenced by avoidance of negative evaluations DEMAND CHARACTERISTICS: cues in experimental setting that inform participants of the purpose and what's expected of them EXPERIMENTER EXPECTANCY: unintentionally bias results by demand characteristics or in their interpretation of data (i.e. researchers errors are liekly to support their hypothesis) CONTROL: use deception, unoptrusive/nonreactive measures, single/double-blind study |
|
|
Multiple Treatment Itnerference
|
Threat to external validity
Order Effects: effects of one level of IV are influenced by previous exposure to different level of IV COUNTERBALANCED DESIGN: different subjects receive IV in different order LATIN SQUARE is a type of counterbalanced desing |
|
|
Between-Group Design
|
different levels of IV are administered to different groups and status on DV is compared betwen groups
|
|
|
Factorial Design
|
Between-Group Design or WIthin-subjects
when study has 2+ IV advantage: more thorough information about the relationship among variables (main effects and interaction effects) |
|
|
main effect
|
effect of one IV on DV
IV-->DV interpret main effects with caution if there is an interaction effect |
|
|
Interaction effect
|
effects of 2+ IV considered together
i.e.: effects of IV are different for different levels of other IV IV+IV-->DV |
|
|
Within-Subjects Design
|
all levels of IV are administered to all participants sequentially
|
|
|
single-group time-series design
|
within-subjects design
measure DV several times at regular intervals before and after IV is administered subjects act as their own no-treatment control (-) susceptable to carryover effects, control through counterbalancing, confounded by autocorrelation (pre-post test correlation inflated because they are correlated b/c same person) - increase Type I error, use special stat to correct |
|
|
Mixed Designs
|
combines between-groups and within-subjects
common when measuring DV over time or across trials (time/trial is considered IV of within-subjects) |
|
|
single-subject design
|
CAN BE USED WITH GROUPS
1)there is a baseline (no tx) phase 2) then there is a tx phase each participant acts as own control 3)DV measured repeatedly during both phases AB design, ABA, ABAB REVERSAL/WITHDRAWAL DESIGN: anytime treatment is administered and then withdrawan. Provides additional information, but may be inappropriate when unethical to remove effective tx MULTIPLE BASELINE: no withdrawal required because you adminsiter tx in different settings (across settings) or to the same bx in different participants (across subjects) |
|
|
Scales of Measurement
|
N: Nominal: Categories
O: Ordinal: ordered numbers, not equal intervals (Likert Scales, race rankings) I: Interval: Ordered and at equal intervals. You can add and subtract to calculate M an SD (IQ, temperature) R: Ratio: ordered,equal intervals and absolute 0, can multiply and divide to determine how much more/less of a characteristic one has over the other (claories, correct answers, response time) |
|
|
mesokurtic
|
normal distribution
|
|
|
leptokurtic
|
more peaked than normal distribution
|
|
|
platykurtic
|
less peaked/flatter than normal distribution
|
|
|
postively skewed distribution
|
most scores are on the low side (tail is on positive side)
Mean>Median>Mode |
|
|
negatively skewed distribution
|
more scores are on the high side (tail is no negative side)
Mode>Median>Mean |
|
|
Measures of central tendency
|
Mode: most frequently occuring number, can be multimodal, easy to identify, but changes with sample fluctuations, not useful for statistical purposes
Median: score that divides distribution in half, if even, it's the number between middle scores, not influenced by outliers, more useful than mode, but only a descriptive stat Mean: average of socres, least susceptable to sample fluctuations, unbiased estimate of population mean (mu), misleading if outliers |
|
|
Range
|
lowest score-highest score=range
can be misleading when there are outliers |
|
|
Variance
|
Mean Square
More thorough measrue of variability includes ALL scores in distribution (not just high and low) sum of (x-M)squared / N-1 average amount of variability in a distribution, indicating the degree scores are dispersed around the mean If calculating population variance, denominator is N If calculating sample variance, denominator is N-1 |
|
|
Standard Deviation
|
square root of the variance
in the same unit as the measurement (variance changes the unit of measurement, making it hard to interpret) The larger the SD, the greater the dispersion of scores around the mean |
|
|
Central Limit Theorum
|
as the size the sample size increases, the sampling distribution of the mean approaches normal
the mean of the sampling distribution of the mean is equal to the population mean the SD of the sampling distribution mean is equal to population SD divided by square root of sample size (standard error of the mean) |
|
|
Sampling distribution of the mean
|
used to determine the probability of a sample having a particulat mean that can be drawn from population with known parameter
the mean of infinite number of samples in the population foundation of inferential statistics |
|
|
Standard error of the mean
|
an estimate of variability that any particular sample may differ from the population
measure of variability due to random error the larger the population SD and smaller the sample size, the larger the standard error |
|
|
null hypothesis
|
stated to imply that the IV has no effect
|
|
|
alternative hypothesis
|
stated to indicate that IV has an effect
may be directional (one-tailed) or nondirectional (two-tailed)...Use Two-tailed unless theoretical grounds to use one-tailed stated in population parameters |
|
|
Rejection Region
|
the values not likely to occur by chance or sampling error.
values lie in the "tail" or the alpha value if value is in this region - null hypothesis is rejected |
|
|
Retention Region
|
value falls in area that is consequence of sampling error only
retain null hypothesis |
|
|
alpha
|
level of signficance
whena two-tailed test, this is divided between the two tails |
|
|
Type I error
|
say there are results when there aren't (rejects true null)
probability of occuring = alpha increased likelihood when small sample size or observations are dependent |
|
|
Type II error
|
retain false null (baby out with bathwater)
probability of occuring = Beta more likely with low alpha level, small sample size, IV not administered at sufficient intensity |
|
|
Power
|
when a test enables experimenter to reject false null (make a correct decision)
maximize by: increase alpha (.05 not .01) increase sample size increase effect size (IV strength and length) minimize error (reliable DV measure, reduce variability within groups, control extraneous) use one-tailed when appropriate Use parametric test (t-test, ANOVA are more powerful) |
|
|
Confidence
|
the certaininty ar esearcher has that the decision they made regarding the null hypothesis is accurate
|
|
|
Statistical test to use with nominal data
|
Chi-Square
-single sample (only 1 variable) -multiple sample (2+ variables) *IV and DV aren't differentiated, both count toward variables* |
|
|
Statistical test to use with Ordinal Data
|
Mann-Whitney (2 independent groups)
Wilcoxon matched-pairs test (2 correlated groups) Kruskal-Wallis test (2+ independent groups) |
|
|
Statistical test to use with Interval/Ratio Data
|
t-test for single sample (sample v. population mean)
t-test for independent samples (2+ independent groups) t-test for correlated samples (2 correlated groups) one-way ANOVA (1 IV, 2+ independent groups factorial ANOVA (2+IVs) repeated measures ANOVA (2+ correlated groups) ANCOVA (removes EV) mixed ANOVA (independent and correlated groups) Randomized block ANOVA (extraneous variable) trend analysis (quantitative IV) MANOVA (2+ DVs) |
|
|
Parametric Tests
|
Interval or Ratio
Normal Distribution homoscedasticity - variance of the popoulations represented are about equal |
|
|
Robustness
|
equal numbers in group
large sample size low alpha level |
|
|
Nonparametric Test
|
nominal or ordinal data
"distribution-free" tests less powerful |
|
|
critical value
|
number that corresponds to the boundary between rejection and retention region.
determined by alpha and degrees of freedom |
|
|
degrees of freedom
|
number of values that are "free to vary" based on the numbers that are already known
calculations: t-test: df=N-1 chi-square df= ("columns"-1)(rows-1) |
|
|
Chi-Square Test
|
analyzes frequency of observation in each category
determine if observed frequency is equivalent to expected frequency (expected is null hypothesis i.e. no difference) expected frequency for each category must be more than 5 and each observation can only appear in 1 category |
|
|
Single Sample Chi Square
|
"goodness of fit model"
statistic: x squared df: (c-1) |
|
|
Multiple sample chi square
|
2+ variables
df= (c-1)(r-1) |
|
|
Mann-Whitney U Test
|
1 IV, 2 independent groups
1 DV, ordinal |
|
|
Wilcoxon Matched-Pairs Signed-Ranks Test
|
1 IV, 2 correlated groups
1DV, ordinal |
|
|
Kruskal-Wallis Test
|
1 IV, 2+ independet groups
1 DV, Ordinal |
|
|
t-test for single sample
|
1 IV, single group
1 DV, interval/ratio df: (N-1) |
|
|
t-test for independent samples
|
1 IV, 2 independent groups
1 DV, interval/ratio df= (N-2) where N = number of subjects |
|
|
t-test for correlated samples
|
1 IV, 2 correlated groups
1 DV, interval/ratio df = (N-1) where N=number of PAIRS of scores |
|
|
one-way ANOVA
|
1 IV, 2+ independent groups
1 DV, interval/ratio df = (C-1), (N-1) where C = number of IV levels and N = number of subjects F ratio: treatment + error / error, the larger F the more likely to have Tx effect. Post hoc needed to determine which means were significantly different |
|
|
Factorial ANOVA
|
2+IV, independent groups
1 DV, interval/ratio F statistic |
|
|
Randomized block ANOVA
|
use extraneous variable as IV
reduces within group variability increases statistical power |
|
|
ANCOVA
|
control extraneous variable by statistically removing it
reduces within-group variability increases power |
|
|
repeated measures ANOVA
|
for within-subjects design
|
|
|
mixed (split-plot) ANOVA
|
used formized design
|
|
|
Trend Analysis
|
1+ quantitative variablea nd wants to know shape of relationship on DV (linear or nonlinear)
|
|
|
MANOVA
|
1+ IV, independent groups
2+ DV, interval ratio simultatneously assessing effects of IV on all DVs (increases statistical power and control experimentwise error) |
|
|
what measures effect size?
|
Cohen's d
r squared and eta square |
|
|
Cohen's d
|
measure of difference between 2 groups
Mean(group A)- Mean(group B) and difvide by SD for both groups gives you a difference in SD units small = .2 medium = .5 large = .8 |
|
|
r square and eta square
|
effect size measurement
indicate percent of variance in outcome variable that's accounted for by tx |
|
|
correlation coefficient
|
summarizes the degree of associateion between variables with one number
|
|
|
pearson r *
|
most common with interval/ratio
|
|
|
Spearman rank-order (Rho)
|
2 ordinal variables
|
|
|
Phi
|
2 nominal variables that are both tru dichotomies (i.e. male/female vs. fiction/nonfiction)
|
|
|
Tetrachoric
|
2 nominal variables, BUT artificial (favorable/unfavorable, successfull/unsuccessful)
|
|
|
Contingency
|
2 nominal variables
|
|
|
Point Biserial *
|
one true dichotomy (nominal)
one interval/ratio |
|
|
Biserial *
|
Artificial dichotomy (nominal)
Interval/ratio |
|
|
Eta *
|
interval/ratio vs. interval/ratio
DOES NOT NEED TO BE LINEAR RELATIONSHIP |
|
|
Assumptions of correlation coefficient
|
Linearity
Unrestricted Range for both variables (i.e. heterogenous groups) - if restricted range, pearson r is underestimate Homoscedasticity: the range of Y scores is about the same as range of X scores - will result in correlation that does not represent a full range of scores |
|
|
coefficient of determination
|
square the correlation coefficient to determine shared variability
ONLY square a correlation if it correlated to different variables |
|
|
Regression Analysis
|
allows predictions to be made
ASSUMPTIONS: linear relationship degree of predictive ability is directly related to magnitude of correlation coefficient standard error of estimate is used to construct confidence interval of predicted score |
|
|
Multiple Regression
|
2+ continuous or discrete predictors used to predict single continues criterion
R (multple correlation coefficient) indicates degree of association between criterion and predictors ideal to have high correlation with criterion and low correlation wtih other predictors multicolinearity: when predictors are highly correlated simple or stepwise |
|
|
multicolinearity
|
when predictors (multple regression) are highly correlated
|
|
|
Types of multiple regression
|
simple: analyze all predictors on criterion at once
stepwise: explain greatest variability with fewest predictors -forward stepwise: one predictor added to analysis -backward stepwise: start with all predictors and remove one at a time |
|
|
Multiple Regression vs. ANOVA
|
Multple Regression: when groups are unequal size, IV is continuous scale, allows researcher to add or subtract IV to determine the best model
|
|
|
Cross-Validation
|
try it out on another sample
likely to experience "shrinkage" because doesn't fit exactly like original sample |
|
|
canonical correlation
|
2+ continuous predictors to predict 2+ continuous criteria
Identifyt he number and nature of underlying dimensinos that account for the correlation between 2 sets of variables |
|
|
Discriminant Function analysis
|
2+ continusou predictors to estmate status on 1 nominal criteria
accuracy measured by "hit rate" or accuratly classified cases |
|
|
Logistic Regression
|
predict status on discrete criteiorn using 2+ continuous or discrete predictors
REQUIRES LINEAR RELATIONSHIP (unlike discriminant function analysis) |
|
|
Path Analysis
|
extension of multiple regression
translate theory about causal relationships into diagram indicating direction and strength between variables |
|
|
LISREL
|
when causal model includes recursive (one-way) and non-recursive (two-way) paths
observed variables AND latent traits variables are believed to measure |
