Quant Midterm

Question 0

Baron & Kenney: 4 statistical criteria for mediation?

Answer 0

1. Predictor variable significantly associated w outcome variable
2. Predictor variable significantly associated w mediator
3. Mediator significantly associated w outcome variable
4. Previously significant relationship btwn predictor and outcome diminished once effects of mediator are controlled
   Can be tested w regression equations

Question 1

Define mediation

Answer 1

Explanatory link in relationship between 2 other variables

How, or by what means, does x affect y through one or more intervening variables, m

Question 2

Define moderation

Answer 2

Factor that gives info about under why conditions and for whom the main effect exists. A factor that may alter the main effect

Question 3

Two main types of sampling?

Answer 3

Probability sampling: occurs when sample is selected from the population w each member having equal probability of being included in the sample.

Convenience (non probability) sampling: chosen for ease of access

Question 4

Quota sampling

Answer 4

Non probability sample in which researcher selects people according to a fixed quota. Aims to represent major characteristics of population by sampling proportional amount of each.

Question 5

Clinicians illusion

Answer 5

Inherent confounding variable is duration of illness: clinicians sample is skewed right

Clinicians see prevalence (those currently suffering) vs incidence (first onset)

Not representative of full range of disease, bc clinical settings tend to comprise people w longer duration of illness and repeated episodes

Question 6

Baron & Kenney approach to mediation an moderation

Answer 6

Moderation: m is a moderator if m explains under which conditions a is related to o (ie there is a significant interaction btwn m and a)

Mediation: m is a mediator if it helps explain how or why a is related to o (assumes no interaction btwn a and o)

Question 7

MacArthur approach to mediation and moderation

Answer 7

Moderation: sig interaction btwn m and a. m must precede a and m and o must be independent (moderator cannot be associated w outcome)

Mediation: m must precede a. Sig interaction btwn m and a OR main effect of m without association btwn m and a.

Question 8

Baron & Kenney vs. MacArthur

Answer 8

B&k do not assume interaction in mediation. Assuming interaction is zero may cause part of interaction effect to be remapped into main effects or error, increasing probability of type I or ii error. Also assumes causal pathways are known. Theory informs identification of mediators vs moderators.

MacArthur requires temporal precedence to make causal inference. Interaction term is always included: bc obvservational methods cannot assume causality, you cannot know whether a variable is a mediator or moderator. Bc of this, approach cannot be used in cross sectional or retrospective atudies

Question 9

Efficacy studies

Answer 9

Higher internal validity

Focus on measuring specific effects of a specific intervention

Question 10

Effectiveness studies

Answer 10

Higher external validity

Focus on generalizability: will this treatment work across settings?

Question 11

Experimental studies

Answer 11

Direct manipulation of iv presumed to be causal agent
Assessment of an outcome thought to directly vary as a function of that iv (causal agent)
Some ethos of inferring what outcome could be wo causal agent (control)

Question 12

Why randomize?

Answer 12

Joint source population: population from which you are inferring results about dv is theoretically the same in all conditions
Extraneous variables not directly measured (confounders) cannot be the cause of a sig relationship btwn x and y bc they should be equally distributed among all conditions

Question 13

What aspect of experimental design allows us to say x is a sufficient cause of y?

Answer 13

Temporal relationship: deliberate manipulation of x means it came before y and was sufficient to cause y

Question 14

What aspect of experimental design allows us to say x was necessary cause of y?

Answer 14

Control group of condition: since control did not get y, x was necessary.

Question 15

How do you address randomization in a within subjects design?

Answer 15

Counterbalancing: by randomizing orders of condition, you minimize ordering effects. Order cannot be the reason x is related to y

Question 16

Define a quasi experimental study

Answer 16

Causal inference vs causality
Like randomized experiments, but wo full control to randomize participants to exposure and non exposure conditions
Certain things cannot be manipulated by the researcher but can still happen in time and space and be captured in a study

Question 17

Cohort vs case control studies

Answer 17

Cohort: selection of participants based on exposure status (iv)

Case control: selection based on outcome variable (DV)

Question 18

Prospective cohort study

Answer 18

Exposure status obtained and measured at start. Group followed in time to ascertain disorder or outcome of interest.
Useful when interested in developmental timing of risk.
Limits recall bias

Question 19

Retrospective cohort study

Answer 19

Group identified based on some exposure that occurred in past, then outcome is ascertained.
All history is traced in past–can be done w archival data.

Question 20

Car control study

Answer 20

Quasi experimental study in which selection criteria is based on outcome variable, ie disease. Controls are wo disease.
Proportion of those exposed to risk factor should be higher in cases than controls.
Want newly diagnosed cases to avoid clinicians illusion.
Control group should be obtained independent of exposure status, should have equal chance of exposure risk.
Confounding can be dealt w statistically, through matching

Question 21

Internal validity

Answer 21

Extent to which study minimizes systematic error (biases in measurement)

Threats include:
History
Maturation
Experimenter expectancy
Subject demand characteristics