Uncertainty Management Flashcards
1
Q
UM Flow Chart
A
- UNCERTAINTY = Methodological/Statistical
- METHODOLOGICAL = Internal/External (RM)
- STATISTICAL = Descriptive/Inferential (STAT)
2
Q
UM in Research
A
- Psych research confronts the unknown/uncertain
- METHODOLOGICAL = how does theory basis/design/data collection reduce uncertainty?
- STATISTICAL = how do stat method analysis reduce uncertainty?
3
Q
Methodological Uncertainty
A
- Related to confidence that design/procedures allow for question to be answered; generally, we try to reduce this considering these sources:
- INTERNAL = what else might explain findings?
- EXTERNAL = to what extent are they true beyond this study?
4
Q
Internal Uncertainty
A
- Not all challenges can usually be resolved in 1 study; surveys/quasi-experiments leave residual IU about causal relationship between correlated variables
- Appropriate experiments address this.
5
Q
External Uncertainty
A
- Relates to researchers ability to generalise findings to non-experimental contexts.
- Most researchers tackle the same theory w/multiple methodology strategies, hence why papers report multiple studies.
6
Q
Reducing MU
A
- Choosing appropriate method
- Eliminating confounds/alternatives
- Controlling extraneous variables (“noise”)
- Controlling internal validity threat
- Representative samples
7
Q
MU VS M-Validity
A
- Validity is a study quality and doesn’t change; MU refers to consumers so can change w/time/understanding.
- Psych debates affect MU retrospectively, not the validity of the time.
- MU can be reduced w/appropriate research practice, though hard to quantify so no agreed method.
8
Q
Statistical Uncertainty
A
- Opposite of MU; design/data don’t minimise it; there are methods to do so but are difficult; stats measures it.
- DESCRIPTIVE (not all pps give same response) or INFERENTIAL (not confident that only the IV affects DV)
9
Q
Descriptive Uncertanity
A
- When psychologists look at phenomenon, they make multiple observations which make multiple data batches.
- 20 pps will react differently to separation from a loved one; even studying 1 pp gives different reactions under various conditions; a sole description always has to be in doubt, expressed in stats via dispersion of central tendency; DU increases w/SD and variation.
10
Q
DU Sources
A
- Inter-individual differences (variation between pps)
- Intra-individual differences (variation within pps (over time))
- Measurement error (variation due to inconsistent measurement
11
Q
Inferential Uncertainty
A
- Research doesn’t just want to comment on the pps; it wants inferences on pops based on sample data (inferential stats).
- Affected by: SIGNAL (sample beh) NOISE (random variance) SAMPLE SIZE; all contribute to p-value; the higher it is, the more uncertainty there is about “genuine” effect not via randomisation
12
Q
Probability Statements (P-Values)
A
- How likely behaviour is due to randomisation.
- If the outcome isn’t due to randomisation (mind the design here), the phenomenon is the only explanation.
- Low prob = less IU and more confidence that the phenomenon is genuine (not a fluke).
13
Q
“Reducing” IU
A
- IU is more measured; if it was eradicated, the phenomenon was likely trivial as it would involve mundane fact (ie. every human has a brain) so a little IU gives the research weight
14
Q
STATSIG VS PSYCHSIG
A
- STATSIG (low IU) ISN’T PSYCHSIG; just because two means are STATSIG doesn’t mean they’re psychologically interesting
- hypothesis system questioned as STATSIG work prioritised, leading to trivial papers published
- “effect size” is common now to show irl importance of work
15
Q
Q1: How should uncertainty be measured?
A
- Practice always involves trade-offs/compromises; being sure about some things increases uncertainty in others.
- Eg. Reducing DU w/homogeneous sample increases EU of broader applicability; this is done via measurement of sensitivity/accuracy; relevance may be reduced since IRL beh is affected by extraneous variables so design is unrealistic; reduces ecological validity.
