Non-experimental designs

Question 1

Quasi-experimental design

Answer 1

Do NOT have complete control over variables
Can NOT randomly assign persons to level of IV
-> non-equivalent control groups
Research in applied settings (NOT LAB)
Must evaluate/minimize threats to internal validity
Must provide treatment to all

Increases external validity
Threats to internal validity remain
Design allows to estimate and minimize some threats

CANNOT IF

• unethical to withdrawal treatment
• treatment causes a lasting change

Question 2

Non-equivalent control groups design with pretest and posttest

Answer 2

Groups already exist it world.
Groups receive same pretest and posttest, only difference, experimental groups receives treatment

If groups equal on pretest, then selection and regression to mean minimized

If no change in control group between pre and post:
-history, maturation, and testing are less likely threats

Mortality and interactions with selection are possible threats

Question 3

Proxy pretest

Answer 3

Differs from posttest, but should correlate with it

Minimizes testing as threat to design

Question 4

Matching in non-equiv. control groups design with pre and post

Answer 4

Match subjects on each group on pretest scores (drop lowest scores of lower mean, and highest scores of higher mean)

• try to make groups more equivalent
• can enhance regression
• > will appear that successful treatment failed!

Selection as threat
-> matching is dangerous with non equiv. groups

Question 5

Separate pre and post samples

Answer 5

Split exper and control groups in half, randomly- each half gets wither pretest or posttest, not both

Eliminates testing threat, only if large number (equivalent in random assignment)

Question 6

Interrupted time series

Answer 6

One group of subjects
Multiple assessments before and after treatment

Evaluates trends in performance before treatment, trend sharply changes only after IV, and see lasting length of trend

Eliminates selection/interactions with selection
Mortality/testing/maturation: can be estimated, if constant
Regression: unlikely if pre-treat scores not extreme and didn’t vary much
History: reduced likelihood if trends are stable pre and post-treat

Question 7

Interrupted time series with controls

Answer 7

Interrupted time series, with control group

Question 8

Interrupted time series with removed treatment (reversal)

Answer 8

When want to admin treatment, and treatment has true effect (effect will reverse when treat removed)

Question 9

Experimental (between-subjects) VS Quasi-experimental design

Answer 9

1. How groups are formed
   E: by experimenter
   Q: existing in world
2. Can an experimental study include a control group?
   E: control and treatment group by random assignment
   Q: not random assignment
3. Pretest
   E: given before groups are formed, in order to match
   Q: given after groups formed

Question 10

Hawthorn effect

Answer 10

Being watched changes behavior