Quantitative methods(02) - The Scientific Method
-
Empirical Cycle
graph TB A[Observation]-->B[Hypothesis] B-->C[experiments] C-->D[Analysis&Result] D-->E[Evaluation: Hypothes supported, adjusted, or rejected] E-->A
-
Observation
- function: sparks idea for hypothesis(anything we want to explain):
- pattern
- unexpected event
- interesting relation
- source: not important:
- personal
- shared
- imagined
- previous research findings
- observing relation in one or more instances
- function: sparks idea for hypothesis(anything we want to explain):
-
Induction
graph TB A[mutiple observation: true in specific cases] --> B[general rule: ture in all cases ]
- general rule or hypothesis
-
Deduction
- hold true in new instances
- expectation/prediction is deduced about new observations
- determine research setup
- define concepts, measurement instruments, procedures, sample
- hypothesis is transformed with deductive reasoning and specification of research setup into prediction about new observations
-
Testing
- data collection
- compare data to prediction
- statistical processing
- descriptive: summarize
- inferential: decide
-
Evaluation
- Conclusion:
- Results confirm prediction–> Hypothesis provisionally supported
- Results disconfirm prediction–> Hypothesis not automatically rejected:
- repeat with better research setup(empirical cycle)
- adjust hypothesis
- reject hypothesis
- Conclusion:
-
-
Confirmation/Disconfirmation
-
Confirmation
- Prediction confirmed
- Hypothesis not proven but provisionally supported
- more support–>more credence
No scientific empirical statement can ever be proven once and for all.
-
Disconfirmation
- Prediction disconfirmed
- Plausible explanations for failure:
- methodological issues
- research design
- instruments inappropriate
- variables not controlled
- Plausible explanations for failure:
- Hypothesis
- preserved
- reject auxiliary assumptions on research design and measurement
- come up with a better research setup
- adjusted
- by adding additional clauses
- modified hypothesis less general
- rejected/radically adjusted
- very rare
- preserved
- Prediction disconfirmed
-
-
Criteria for evaluation
-
Reliability
- Replicability: repetition possible
-
Validity
hypothesized relation accurately reflects reality
-
Construct validity
- prerequisite
- constructs measured/manipulated accurately
- instruments measure/manipulate intended properties accurately
- biggest challenges!!!
-
Internal validity
- hypothesized causal relation
- observed effect due to hypothesized cause
- threatened by: plausible alternative explanations
-
External validity
- hypothesized relation holds in general?
- results generalize to other people,groups,environments,times?
-
-
-
Causality
- Hume first list the criteria:
- cause and effect are connected
- cause precedes effect
- cause and effect co-vary consistently
- correlation does not imply causation!
- no alternative explanations
- the most difficult one!!!
- methodology to minimize the alternatives
- Hume first list the criteria:
-
Internal Validity Threats
-
Participants
-
Maturation
- alternative explanation: natural change
- solution: control group
-
Selection
- alternative explanation: differences in participant characteristics
- solution: randomized selection
-
Selection by maturation
- alternative explanation: groups systematically differ in their rate of maturation
- selected groups become mature during the experiment
- solution: randomized assignment to groups
-
-
Instruments
-
Low construct validity
- systematic bias/measures another construct
- prerequisite for internal validity
- solution:
- valid instruments
- valid manipulation methods
- used consistently
-
Instrumentation
- instruments changed during the process of the study
- solution:
- valid instruments
- valid manipulation methods
- used consistently
-
Testing
- sensitization: measurement affects behavior
- solution:
- special design with multiple groups with/without pretest
-
-
Artificiality
expectation changes behaviors
-
Experimenter expectancy
- Researcher changes behavior(unconsciously) biasing effect researcher’s expectations
- influences participants’ responses
- solution: experimenter blind design
-
Demand characteristics
-
Participant changes behavior(unconsciously)
biasing effect participant’s expectations
-
Solution: leave participants unaware of the real purpose of the study or at least which group they are in
- Double Blind Research Design
-
solution: cover story
- temporary deception–>needs to be necessary
- risk of bias–>needs to be real
- researcher–> needs to debrief participants afterwards
-
-
-
Research Setup
-
Ambiguous temporal precedence 时间先后模糊
-
solution: manipulation of cause
-
History
- unforeseen event during study
- provides alternative explanation
- large-scale
- solution: unavoidable
- small-scale - solution: test subjects separately, if possible
- unforeseen event during study
-
-
Mortality
- participants dropout
- alternative explanation: differential dropout
- solution: hard to avoid; document reasons for dropout for further research
- participants dropout
-
-
Relevant types of variables
-
Construct
- denotes property in general, abstract terms
- loneliness&depression
- denotes property in general, abstract terms
-
Variable
-
operationalized, concrete expression of construct.
- measurable/manipulable
- UNLA loneliness scale/ GDS depression survey
-
Vary+able!!
-
Types of variables
-
Independent variable
- In control of
- cause; explanatory; input; predictor
-
Dependant variable
- influenced by/the result of dependent
- effect; response; outcome; output
-
-
Variables of disinterest
extraneous properties
-
Confounders/ lurking variable
潜在变数(Lurking variable):指对研究中其他变数间的关系有重要影响,却没有被列入研究范围的变数.(要么因为此存在变量不为人知,或者它的影响被认为是可以忽略而其实不能忽略,或者是数据无法获得)
- related to independent/dependent variables
- partially or entirely accounts for relationship
- solution: keeping it constant or turn it into control variable
-
Control variables
- likely to be related to independent/dependent variables
- effects can be controlled for (unlike confounder)
- solution: checking relationship at each level or value
-
Background variable
- not relevant in relation between independent+dependent variable
- relevant for determining representativeness:
- age
- gender
- ethnic/cultural background
- social status
- education level
- solution: change it into control variables
- used to assess generalizability
-
-
-