Please enable JavaScript.

Coggle requires JavaScript to display documents.

Task3 (GARB (RESULTS OF STUDY (SUMMARY (Researchers constructing a…

- - - -  PVP (positive predictive value of the positve scores): calculated by identifying subset of trainees who gave less desirable response and calculate proportion of them to trainees discharged within follow-up period
      - – PVN (predictive value of negative scores): quantity of an item calculated by identifying the subset of trainees who gave most desirable answer and calculate proportion of those to trainees discharged in follow-up
  - - - Researchers constructing a predictor scale in applied setting should consider:
        
        o External strategy with empirically derived or unit weights
        
        o Internal with unit weights
        
        o Intuitive with unit weights
      - When predictor and outcome data are available and derivation sample is large  external strategy best choice (with regression weights)
      - Predictor but not outcome data and sample large enough for factor analysis  internal strategy with unit weights
      - Sample size sufficiently small: advantage of empirical weights and external approach may be of negligible practical importance
      - External validity - used for heterogeneous items - better for predicting, high criterion validity (e.g. for testing interest) (prof strong from course manual)
      - Ionternal method- to ttest homogenous items - less predictive - weak criterion vliadity
- - - - expert rankings (measure of deep coma in patients)
      - method of equal-appearing intervals (dichotomous items judged relevance by group of experts)
      - Method of absolute scaling (also thurstone) - procedure for obtaining absolute difficulty of itemsbased on age groups of test takers (items distributed to 2 groups, where one age group serves as an anhor, scores are compared and SD from anchor group mean is becoming an item difficulty)
      - Guttman scales
      - Likert scales (1-5)
      - Method of empiricla keying - items based purely on statistics
      - rational scale construction (internal consistency - all items correlate positively with each other and with total score for the scale)
      - thurstone scaling approach- randomized items rated i respect to severity (mild-low depression)
    - - short answer objective form
      - matching questions (drawin lines)
      - Multiple choice
      - Foirced-choice methodology
      - individualy administered tests
    - - item reliability index = poin biserial correlation= correlation between item score and total score
      - item validity index = SD * reliability
      - item difficulty = p value 0.3-0.7
      - item characteristic curve
        
        index of item discrimination = slope on ICC
- - - - internal
        
        Homogenity
        
        inetrnal consistency approach - you try to create as consistent scale as possible - data driven method - search for items that are already created and put them together
        
        scale is created through factor analysis
        
        cross validation asa validation method
        
        especially good when we dont want to waste time
      - external
        
        criterion validity
        
        criterion-oriented method = all about prediction. Completely data driven. (prof. strong from course manual)
        
        example = test to measure interests
        
        look for heterogeneous scales = purel based on correlation between items and criterion
        
        Cronbachs alpha is unsuitable = becasue no consistency
        
        best used for predictions of non-theoretical criterion
    - - Construct
        
        Construct validity
        
        Nomological network (specify what are possible relations betwen different factors-mediators, moderators etc.), maximizing construct validity. Needs strong theoretical basis.
        
        convergent and discriminant validity have to be specified in advance
        
        multi-trait multi-method matrix
        
        there must be already lots of research in order to create good theoretical basis. Its very expensive and time consuming
        
        items are defined by theory
      - Facet
        
        Content validity
        
        similar to construct but doesnt need nomoogical network. aimed at maximizing content validity but is still theory-driven and the construct needs to be welel defined (the focus stil on construct) but its not as frequently used
        
        creation of homogeneuos facets of a construct
        
        Focus on defining construct
        
        good when you need to deifine ocnstruct with different elements
    - - rational
        
        Face validity
        
        based on expert opinions (comon sense method)
        
        example - clinicj judgments, composite questionaires for research
        
        Based on vision/general idea. Based omn it the idea how to conduct anlaysis is created. Then it is debated which items are relevant and apropriate itemsa re created. Ratings are judgeb by experts, and scale is sufficient if everyone agrees so.
        
        goog for global/general problems
        
        no gneeral framework
      - prototypical
        
        Process validity (the behaviours elicited in a given test are representative of given dimension)
        
        categories of constructs are of continuum nature - boundaries between constructs are vague
        
        Example - personality, impulsivity questionaire
        
        no framework, or construct analysis or specification f items.
        
        1st = ask teachers what is the most impulsive chld in their class like. 2nd collect al the ratings and give proper weights (act frequency approach). u create scale out of it
        
        good for constructs that are hard to define (f.e. personailty)
        
        Limitaiton = no theoretical framework