Exam 4 Flashcards
Field Research
Naturalistic observation, Case studies, Archival
Qualities of Field research
little or no manipulation, little or no random assignment, less control
Whats the point of field research
External validity, easy generalization to real world
When do we use a Quasi-Experiment?
When a true experiment is not possible, unethical to move people around (like sick/not), and it is not possible to change peoples genes
Qualities/aspects of a quasi-experimental design
causal hypothesis, at least 2 levels of an IV(not always manipulated), specific procedures for testing hyp, some controls for threats to validity(double-blind,automation etc.)
Types of quasi-experimental designs:
Non-equivalent control-group design
Interrupted time-series design
Non-equivalent control group design
pre-existing control/experimental group, groups can be made similar on important
Interrupted time-series design
One group tested repeatedly, Within subjects design
Types of designs for Program Evaluations:
randomized control group design, non-equivalent control group design, single group interrupted time series, pretest-posttest design(w/no control group, not recommended)
Why transfer data?
Make non-normal distributions, or conceptual reasons for better understanding of the data
Types of transformations
sqrt(X), Log2(X), 1/(X)
Data transformation:Skew, how to report?
Use the raw data for: Descriptive summary values (Mean, SD, N)
Use Transformed data to: Run the parametric tests(t,F,ANOVA)
Platykurtotic
flat (values distributed evenly)
Leptokurtotic
Tall (values mostly around the mean, and less extremes)
Types of data
Nominal, Ordinal, Interval, Ratio
Nominal
Data with an order
Ordinal
Ordered, but not necessarily evenly spaced
Interval
Equal interval, no absolute 0
What do you use when you do not have an equal interval but want one?
Item response theory
Describe item response theory
How likely is it that a given person will get a question correct, It put persons responding, and the items they are responding to on the same scale… (so equal interval)
Ratio
Ordered, equal intervals, absolute zero
When do you use the Rasch Model?
For dichotomous responses (like true false)
When do you use the Rasch credit model?
For polytomous responses
Information received from Item Response Theory:
item difficulty (b), Step difficulty (o), and person scores(0)
In multiple regression what is this used for:
Y’=a+b1X1+b2X2
To predict a DV from multiple IV’s (1 DV, IV1, IV2, IV3…etc.)
(allows one to assess how many (or few) IVs predict a DV in a model)
Y’=a+b1X1+b2X2
a = regression constant (intercept) b1 = partial regression coefficient for IV predictor 1 b2 = partial regression coefficient for IV predictor 2 X1 = score on IV predictor 1 X2 = score on IV predictor 2
Hierarchical Multiple regression
Planned, incremental, multiple regression is done in steps, and planned by YOU
Stepwise Multiple regression
Unplanned, incremental, computer selects best IV correlated with the DV successively, Examine R2 and its change. (not the best bc you can end up with a worse model than if you made it)
Issues with Multiple regression:
IV’s may be correlated
multicollinearity
When the IV’s overlap too much
Special correlations
Partial regression coefficient, Semi-partial (“Part”) correlation, partial correlation
Partial regression Coefficients
a regression weight that adjusts for the other regression weights in a multiple regression model, but when both predictors are taken together in the same model, the regression weights change
Semi-partial (“Part”) correlation
when squared, (sr2) is the unique proportion of Y variance uniquely explained by X1
Partial correlation
when squared, (pr2) is the proportion of variance in Y not associated with X2 that is associated with X1
Multiple regression becomes complex when…
you add in IV predictors that are related to each other
Factor Analysis
used to reduce the several variables into sets of variables
assess how well items or scores align themselves on single or multiple dimensions
Two types of Factor Analysis
Exploratory, and confirmatory
Exploratory factor analysis
the goal is to extract a common variance of the variables with their factors
What question does EFA answer?
how many factors are in this set of variables
Factor loading=
correlation of variable X with a factor
Confirmatory factor analysis (CFA)
We know how many factors we want to extract, and we know what relationship the factors should have with the variables
Reliability
consistency or stability in measurement
If a test is not “reliable”…
there is too much measurement error
X (observed score) =
T (true score) + E (Error)
Ways to test reliability:
Test-retest Reliability, Slipt half, Chronbach’s alpha, Item-total correlation
Test-retest reliability
measure the same people twice on the same scale, then compute the coefficient between the two occasions
Why can test-retest reliability be misleading?
scores can shift up, yet maintain the same order, did they improve? or are they stable? ??????
Split half (reliability)
split the test in half and correlate the two halves
Split half strengths
can get reliability estimate with only one test
Split half weaknesses
measuring the same traits throughout the scale?
Chronbach’s Alpha
Average of all possible ways of splitting a test
Problems with Chronbach’s Alpha
the more items there are, the more reliable the test is…which means Cronbach’s alpha is measuring…?
Item-total correlation
take item 1’s score and correlate it with the sum of all others. (If persons score high on this item, and high on the total test, then the relationship should be high.)