Statistics Flashcards
Chi Square
is non-parametric test that does not require normality, homoscedasticity, requires independence of observations so you cannot measure subjects twice or do any kind of repeated measures.
Kappa Coefficient
interrater reliability - the degree of agreement between raters on an instrument that is subjectively scored.
discriminant analysis
several independent variables are used to predict group membership. predicting group membership based on a cluster of independent variables.
multiple regression
several independent variables used to predict one continuous dependent variable (i.e., EPPP score)
Cluster Analysis
deriving several subgroups from a cluster of dependent variables (ex. analyzing MMPI data on police officers and finding that police officers fall into three groups)
Manova
Used to analyze the effects of one or more IV on two or more DV. Comparing several groups on several DVs. Used when the study includes two or more dependent variables.
1+ IV Effects on 2+ DV
LISREL, Structural Equation Modeling
Stands for LInear Structural RELations test a causal model of relationships among variables only linear relationships and cannot derive causal relationships among variables.
Purpose of Rotation in Factor Analysis
Rotation alters the factor loadings for each variable and the eigenvalue for each factor.
Solomon Four-Group Design
is a true experimental design that is used to evaluate the effects of pre-testing.
Experiment-wise Error rate
the probability of at least one Type I error in comparison experiments.
Time-series Design
take a number of measurements over time and then somewhere along that time you introduce experimental manipulation and if you see change, its due to your manipulation. Biggest threat to internal validity is HISTORICAL EVENT.
Central Limit Theorm
As sample size increases, the shape of a sampleing distribution of means becomes more normal.
Item Response Curve
Provides 3 pieces of information about a test item:
1) its difficulty
2) its ability to discriminate between high and low scorers
3) probability of answering the item correctly just by guessing.
Item Response Theory
mathmatical approach to item analysis
1) characteristics of items should be the same for all theoretically equivalent groups of subjects choosen from the same population. Have been applied to the development of culture-free tests.
2) works well with large samples
3) test items measure “latent traits”
4) useful in development of of computer programs tailored to individual’s level of ability.
Factorial Anova
Used when study has more than one IV
Allows for assessment of both Main effects (effects of each IV considered individually) and interaction effects (effects of each variable at the difference levels of the other variables)
Item Characteristic Curve (ICC)
associated with item response theory, are graphs tht depict individual test items in terms of the percentage of individuals in difference ability groups who answered item correctly. It is costly, but it provides a lot of information regarding individual test itemps including 1) difficulty 2) discriminability, 3) probability that item will be guessed correctly.
True Experimental Research
at least 1 IV is manipulated and subjects are randomly assigned.
ex. study to compare effects of 2 types of tx for anxiety. IV of treatment is manipulated, subjects are randomly assigned to one of the 2 treatment groups.
Quasi-Experimental Research
At least one IV is manipulated but there is non-random assignment because subjects are typically in already assigned groups.
ex. effects of two different tx on patients of Ward A and Ward B of hospital.
Observational, Passive, or Non-experimental Research
no intervention or manipulation.
focus is on statistics are used to detect group differences (ANOVA).
ex. study comparing the smoking in adolescent male and females. In this study IV is gender (non-manipulated IV) and DV is extend of smoking.
Carryover effects
occurs with repeated measures designs in which subjects are exposed to several different interventions, conditions in sequence. Counterbalancing is often used where subjects are divided into thirds and tested in different orders. (i.e. Latin Square is a form of counterbalancing)
Latin Square ex. 123, 312, 231
Between Groups Design
only compares groups that are independent
ex. differences in reading levels of two difference 2nd grade classes.
Within Subject Design
groups contrasted are correlated or related.
1) repeated measures
2) subjects have been matched
3) subjects have inherent relationship (i.e. twins)
Mixed Design
groups that are both independent and correlated.
ex. patients assigned 2 different treatment groups for depression which are later compared. (Between design/Independent groups)
Each group is also measured before and after (Within design/correlated groups or repeated measures)
idiographic
single subject studies where few subjects are studied.
“idiographic” = “individual”
nomothetic
group designs.
Autocorrelation
effect of measuring the same person repeatedly will result in highly correlated data.
AB Design
Single Subject Designs (idiographic)
A (baseline) –> B (treatment)
Threat of history is most problematic = hard to know if the change was due to treatment or some other event.
ABAB Design
Single Subject (Idiographic)
A (baseline) –> B (treatment) –> A (baseline) –> B (treatment)
Whole point is to allow to return to baseline to see if change is really due to tx.
Pro: Helps to protect against threat of History
Cons: failure of DV to return to baseline and ethical issues related to discontinuing effective tx.
Mutliple Baseline Design
Single Subject (Idiographic)
Treatment is applied sequentially or consecutively.
1) Multiple Baseline across subjects = tx in progression with different subjects.
2) Multiple baseline across situations = 1 subject across 3 situations (home, school, work)
3) Multiple baseline across behaviors = 1 subject across different behaviors.
Pro: resolves problems with AB or ABAB designs
Cons: time consuming and expensive.
Simultaneous (Alternating) Treatment Design
two or more interventions implemented concurrently during the tx phase.
ex. alternating between M&M’s and Praise.
Changing Criterion Design
change the criterion in increments.
ex. start with 10 cups of coffee to 8 cups, then 8 cups to 6 cups etc.
Time-Frame Research Designs
1) Cross-sectional research
2) longitudinal research
3) cross-sequential research = combination of cross-sectional and longitudinal
simple random sampling
every member of the pop. has an equal chance of being randomly selected.
Stratified Random Sampling
pop is divided into Strata ( ages levels, income levels, ethnic groups) and then random sample of equal size from each stratum is selected.
ex. selecting equal representation from each SES level (low, middle, upper)
Proportional sampling
individuals are randomly selected in porportion to their representation in the general pop.
ex. pop is 80% white, 10% black, 25% hispanic, sample would reflect same percentages.
Systematic sampling
selecting every kth element after a random start.
ex. if 100 out of 1000 persons are needed, every 10th person is selected.
Cluster Sampling
identifying naturally occuring groups of subjects (clusters) and randomly selecting certain clusters (i.e., classes or departments at university). everyone is that cluster is surveyed and studied.
ex. psychologist wants to study achievement in 8th graders in LA, 10 schools are randomly selected from the district and all 8th graders are assessed from those schools.
Threats to internal validity
factors other than IV that may have caused the change in the DV
1) History
2) Maturation
3) Testing or Test Practice (Solomon Four-Group Design)
4) Instrumentation
5) Statistical Regression
6) Selection Bias
7) Attrition or experimental mortality
8) diffusion
History
Threat to internal validity
specific incidents that intervene between measuring points, either in or outside of the experimental situation
Solution: have a control group
Maturation
Threat to Internal Validity
factors that effect the subject due to the passage of time (fatigue, maturing)
Solution: Control group
Testing or Test Practice
Threat to internal validity
familiarity with test affects scores on repeated testing.
Solution: Solomon Four-Group Design = 4 groups, 1) pre and post with intervention 2) pre and post without intervention 3) post only with intervention 4) post only no intervention.
Solomon Four-group Design
Solution for problem of testing or test practice (threat to internal validity)
divide subjects into 4 groups:
1) pre and post with intervention
2) pre and post without intervention
3) post only with intervention
4) post only no intervention.
Instrumentation
Threat to internal validity
equipment problems
ex. biofeedback electrodes tend to show reductions in scores as equipment wears out.
Solution: Control Group
Statistical Regression or “regression to the mean”
Threat to internal validity
tendency for extreme scores (very much above or very much below mean) to become less extreme (closer to mean) on retesting even without any type of intervention.
ex. high scores on the BDI are likely to score a bit lower second test.
Solution: Control Group
Selection Bias
Threat to internal validity
non-random assignment.
solution: Random assignment.
Attrition or experimental mortality
Threat to internal validity
differential loss of subjects from the groups.
Solution: subjects who drop out should be compared on relevant variables by running t-tests.
Diffusion
Threat to Internal Validity
no treatment group actually gets some of the treatment, clouding treatment effects.
ex. one group receives CBT for anxiety, the other group gets no specific interventions but during research, some CBT was inadvertantly discussed in the control group.
Solution: tigher control over the experimental situation will help.
Construct Validity
factors other than the desired specifics of our intervention that result in differences.
Rosenthal effect
Threats to construct validity
experimenter expectancies are accidently/unconsciously transmitted to subjects by the experimenter through cues, clues.
Solution: having the experimenter blind to the subjects’ treatment condition can control for this.
Demand Characteristics
Threats to construct validity
factors in the procedure that suggest how the subject should behave
Ex. subjects given meds are told of the side effects and the subjects report more side effects. but subjects given placebos are told no side effects and they report none.
Solution: Subjects should be blind to their treatment condition.
John Henry Effect
Threat to Construct Validity
compensatory rivalry occurs when persons in a control group try harder than ususal in the spirit of competition with the experimental group.
Solution: experiemental and control group should not know about each other.
Threats to external validity
factors that threaten generalizability
Sample Characteristics
sampe characteristics not the same as Pop.
Stimulus Characteristics
Threat to external validity
features of the study or artificial research arrangements.
ex. laboratory may not be generalizable to natualistic settings.
Contextual Characteristics
Threats to external validity
conditions in which the intervention is imbedded.
Reactivity = subjects behave in a certain way just because they are participating in research ex. Hawthorne effect = just by nature of being observed subjects behave a certain way.
Threats to Statistical Conclusion Validity
1) Low power
2) unreliability of measures
3) variability in procedures
4) Subject heterogeneity
Intercorrelations of Threats
Greater Internal Validity (control, random assignment) , lower external validity (generalizability)
AKA
Tighter the experimental control, the less it can be generalized to pop.