Please enable JavaScript.

Coggle requires JavaScript to display documents.

SU5- Measurement, Scaling - Coggle Diagram

- - - - Identify measurable subcomponents that can be used to indicate subjective properties
- - - - Ordinal- rank; relative magnitude
      - Internal- absolute differences btw each scale point
      - Ratio
      - Nominal- label
      - Data level determines possible/suitable stats analysis
  - - - Single vs. multiple-item
        
        Single-item- focus on 1 attribute of a construct/object
        
        Multiple-item- focus on multiple attributes of a construct/object
        
        Choice- depends on
        
        Construct- how many dimensions? if construct has several dimensions/subcomponents, to measure each one of them
        
        E.g. service (construct) 5 dimensions (empathy, reliability, responsiveness, assurance, tangibles)
        
        Reliability, validity- multiple items more reliable, valid
      - Wording
        
        Avoid ambuguity
        
        Include all needed instructions
      - Double-barreled e.g. customer service + product performance in same qn
      - Leading e.g. experts believe....do you agree?
      - Loaded- emotionally charged
      - Ambiguous e.g. occasionally + sometimes
      - Complex e.g. adequate (define this instead)
      - Double negative e.g. not incompetent > competent
      - Mutually exclusive e.g. 0-10 + 10-20, use below 10 + 10-20 instead
    - - Non-comparative (without referencing another) formats
        
        Behaviour intention- likelihood of buying/engaging in behaviour of interest e.g. definitely will buy, probably will buy, etc.
        
        Semantic differential- only endpoints labelled with unique, bipolar descriptors e.g. expert, non-expert
        
        Cons- inappropriate scale descriptors e.g. expert, beginner
        
        Likert- extent agree/disagree
      - Comparative (based on some other object) formats
        
        Rank-order e.g. 1st, 2nd, 3rd preference
        
        Paired-comparison- characteristics paired against one another e.g. trust or competence more impt?
        
        Constant sums- allocate a number of points (e.g. 100) among importance of several attribute
    - - Qn understanding (respondents intellectual capacity, language ability)- use simple word choice, sentence; pre-test
      - Scale descriptor discriminatory power- how many scale points needed?
      - Balanced (equal # of +ve/-ve response alternatives) vs. unbalanced (unequal # of +ve/ve response alternative)- use balanced to avoid bias
      - Forced (no neutral descriptor) vs. non-forced choice ("N.A." choice) scale- forced lowers data quality
      - Negatively worded statements- minimise
      - Desired measures of central tendency, dispersion- depends on data level
        
        Mean- avg.(interval, ratio)
        
        Median- divides data into half (ordinal)
        
        Mode- more frequently occurring (nominal)
      - Desired measures of dispersion
        
        Frequency distribution- response count (nominal)
        
        Range (ordinal)
        
        Standard deviation- statistical variation degree (interval, ratio)
    - - Understanding research problem
      - Data requirements
      - Developing constructs- what to measure
      - Appropriate measurement scale- how to measure
  - - - Test-retest technique
        
        Same respondents, different times
        
        Cons
        
        Absent 2nd time
        
        Sensitised, administration differences (environmental/personal factors)- changes responses 2nd time
        
        Different respondents (2 samples), same time/conditions
      - Equivalent form
        
        Internal consistency
        
        Coefficient alpha
        
        Avg of all possible split-half measures (using different ways of splitting)
        
        Good consistency- correlation value of more than 0.7
        
        Split-half test
        
        Divide items into 2 halves
        
        Correlate halves against each other
        
        Good consistency- high correlation
        
        Process
        
        Create 2 different but similar scale
        
        Give both forms to same/different samples
        
        Cons- time-consuming; thus focus on internal consistency instead
    - - Before data collection- qn suitability
        
        Content- extent construct represents all relevant dimensions
        
        Face- based on researcher's intuitive/subjective evaluation
      - After data collection esp. for multi-item scales
        
        Convergent- multiple items measuring same construct have higher variance proportion
        
        Discriminant- extent single construct differs from other construct; unique
      - Data source
        
        Pilot study- respondents representative of target population
        
        Panel of experts- used when limited resources