Please enable JavaScript.

Coggle requires JavaScript to display documents.

Research Methodology (2 Research topic (Value of academic work (Whetten,…

- - - - Positioning
      - Theory
      - Discussion
  - - - Financial TImes
      - UT Dallas
      - Norwegian journal list
      - Association of Business Schools, UK
    - - Substantial problems of IF
        
        inconsistency
        
        2 years is a short period
        
        inflation by journal's self-citations
        
        no distinction between major and minor journals
        
        hard-to-read journals are cited less
        
        variation across articles in same journal
        
        both good and BAD reasons to cite
        
        resons by Baumgartner and Pieters, 2003
        
        Application
        
        Affirmation
        
        Negation
        
        Review
        
        Perfunctory mention
      - 5-year impact factor
      - Article influence (AI)
        
        5-year period
        
        self-citations removed
        
        weighted count of citations (well-cited vs. poorly cited journals)
      - Scimago Journal Rank (SJR)
        
        Inspired by Google Page Rank algorithm
        
        3-yaer period
        
        influence of self-citations is limited
        
        weighted count of citations
    - - Top journals by Harald van Heerde
        
        J of Marketing
        
        J of Marketing Research
        
        J of Consumer Research
        
        Marketing Science
- - - - Conclusions on sample should generalize to populations
      - Characteristics in sample should be similar to population characteristics
    - - enough respondents need to be included
  - - - Focal variable is dummy
        
        Variance = p(1-p) where p is fraction of ones from pretest / past / judgment
        
        error = allowed error margin between sample fraction and population fraction
      - Focal variable is continuous
        
        Rule of thumb: range = 6 * SD
        
        Error = allowed difference between sample mean and population mean
        
        Variance from pretest / past observations / own judgment
      - Z = 1.96 if 95%
        Z = 1.645 if 90%
        Z = 2.575 if 99%
    - - random sampling
      - large population
        
        if population size N < 10*n, then multiply by N/(N+n-1)
    - - sample must be large in order to be precise
      - sample size depends on population size
        (only true if population is small)
  - - - Systematic sampling
        
        every i-th element (after initial randomization) included in sample
        
        +: easy to use (if sampling to be done by others)
      - Stratified sampling
        
        construct homogenuous groups (strata), take random sample within groups
        
        Proportional: sample size is proportional to group size
        
        +: representativeness
        
        -: small groups have small representation
        
        Non-proportional: sample size within group is not proportional to group size
        
        +: small groups represented well
        
        -: total sample is not representative
        
        +: control over group sizes, facilitating comparison across groups
      - Simple random
        
        each element has equal probability to be included in sample
        
        +: sample is good reflection of population (generalizability)
      - Cluster sampling
        (area sampling)
        
        construct groups, sample groups randomly and take ALL group elements
        
        +: combining random sampling with convenience -> low costs
        
        -: opinions within clusters often correlated
      - Extension: multi-stage sampling
        combine two or more probability sampling techniques
    - - Judgmental sampling
        
        Only targeted elements are included in sample
        
        take people with extreme views from which you may learn most
        take experts
      - Quota sampling
        
        Construct groups, take non-random samples proportional to group size
        
        Non-probability counterpart of proportional stratified sampling # :
        
        Similarity: group size is proportional to group size in population
        
        Difference: stratified sampling is based on random sampling within groups & quota sampling is based on convenience sampling within groups
      - Convenience sampling
        
        only most accessible elements are included in sample
        
        quick & dirty: sample is unlikely to be representative
      - Snowball sampling
        
        Participants in sample are asked to invite others who are eligible
        
        +: easy to reach members of target group who are hard to reach otherwise
        +: quick way to get many respondents
        -: people close to each other are likely to be similar -> not representative
  - - - 3) compute summary statistic from sample
      - 4) compare summary statistic to reference distribution
      - 2) define the null hypothesis (no effect)
      - 5) draw conclusion
      - 1) define the research hypotheses (the effect we try to find)
- - - - Cause A and effect B are correlated
      - Cause A precedes effect B
      - No other plausible explanations for correlation between A & B
    - - Reverse causality
      - Selection effect
      - Confound
  - - - Internal
        
        Causality within study is real
        
        Is about causality & Alternative explanations must be eliminated (or controlled for)
        
        Experiments have high internal V
        
        Manipulate cause and observe effect
        
        Controlled setting is clean, reducing alternative explanations
        
        Big threats #
        
        more threats if experiment has pre-test and post-test
        
        Hypothesis guessing: measuring DV in pre-test may change DV in post-test & respondents discover aim of study and adjust answers
        
        Attrition (истощение) effects if participants drop-out selectively during test period
        
        History effects if uncontrolled events take place during test
        
        Subtypes of internal V
        
        Content
        
        whether included variables cover all important dimensions of construct
        
        Criterion
        
        to what extent is measurement scale able to predict future criterion variable
        
        Construct
        
        whether constructs measure what they're supposed to measure
        
        Threats
        
        mono-operation bias (lack of content validity)
        
        construct is ambiguous
        
        hypothesis guessing
        
        Subtypes
        
        Discriminant
        
        valid scales for supposedly UNRELATED constructs should have LOW correlation
        
        Nomologicsl
        
        valid scales for supposedly RELATED constructs should give correlations in ANTICIPATED DIRECTIONS
        
        Convergent
        
        valis scales for SAME construct should be HIGHLY correlated
      - External
        
        Study results generalize beyond specific setting in study
        
        Sample composition allows for generalization of study results
        
        Threat: sample composition is too narrow
        
        only people with certain characteristics included
        
        setting is too specific
        
        moment of study is too specific
      - Ecological
        
        Study setup mimics real life (realism)
        
        Example of high ecological V: Coelho do Vale et al., 2009 - chips consusmptions
  - - - Internal consistency
        
        Consistency of results across items representing the construct
        
        Do items hang together as a set?
        
        Measure: Cronbach's aplha
      - Intra-rater test-retest
        
        Are outcomes from SAME participant the same at different points in time?
      - Inter-rater
        
        Are outcomes from DIFFERENT raters for same task the same?
        
        If not: scale is like ambiguous (or bad raters)
        
        Measure
        
        Hit rate: fraction of items ion which both raters agree
        
        Bad measure: possible that the agreement was due to chance alone (Cohen, 1960)
        
        Most popular chance-corrected measure: Cohen's kappa
        
        (H - Hchance) / (1 - Hchance)
        
        Hchance: внешние суммы-углы + внутренние суммы-углы
        
        Interpretation of kappa by Landis & Koch, 1977
        
        Most popular in marketing: Perrault-Leigh index
        
        probability R that rater assigns item to correct category without guessing
        
        R = sqrt ((H-1/C) * (C/(C-1)))
- - - - if allow another variable that influences DV to change along with the manipulated IV
      - Consequence: manipulation isn't clean
      - Solution: keep situational characteristic identical across groups
    - - when a measure designed to assess a particular construct inadvertently (ненароком) measures something else as well
      - threat to construct validity
    - - if individual differences of participants that affect DV are allowed to covary with the experimental treatment
      - Solution: random assignment to experimental conditions
      - Consequence: groups differ, not homogenuous
    - - Passive control
        
        Control by randomization
        
        Subjects randomly assigned to treatment conditions
        
        +: helps to control many extraneous variables at a time
        (in contrast to e.g. matching) -> best method
        
        -: doesn't guarantee control, just makes it likely
        -: dangerous with small N
      - Active control
        
        Control by matching
        
        compare test units on a set of key background variables before assigning them to treatment conditions
        
        requires time and effort, can only be done for few variables, still variation for other variables
        
        Control by holding constant
        
        Influence of extraneous variable is eliminated by holding it constant across treatment conditions (e.g. same day, same gender)
        
        can help to remove the confound effect, but may limit external validity (unable to generalize)
- - - - variable that affects the direction and/or strength of the relation between IV and DV
      - Test for moderation
        
        Y = a + bX + cZ + dXZ (d needs to be significant)
    - - variable representing the generative mechanism through which IV is able to influence DV
      - Conditions for mediation by Barron & Kenny, 1986
        
        2) Mediator Z is explained by IV X
        
        Z = a2 + b2*X
        
        3) DV Y is explained by mediator Z when controlling for X
        
        Y = a3 + b3 X + c3 Z
        
        1) DV Y is explained by IV X
        
        Y = a1 + b1*X
        
        3b) Effect of X on Y should be smaller in absolute value than in step 1
        
        Y = a3 + b3 X + c3 Z (b3 = 0)
        
        Complete mediation if effect of X on Y has become 0
    - - Inverted U-shape
        
        More advertising leads to higher sales -> via awareness
        However, more advertising also leads to lower sales -> via annoyance
    - - Moderator
        
        WHEN does relation between X and Y occur? (conditions)
      - Mediator
        
        HOW does relation between X and Y occur? (mechanism)
- - - - measure constructs and connections
    - - manipulate conditions within groups
    - - joint evaluations of product attributes
      - Research question: which attributes of product or service does respondent consider most important? And which attribute levels are most attractive?
      - Approach:
        
        ask respondents to evaluate hypothetical products
        
        products are combinations of attribute levels composed by researcher
      - Why conjoint analysis instead of a survey about attribute importance?
        
        tradeoffs: cannot say all attributes are important
        
        implicit: no problem of socially desirable answers
        
        realism: assess entire products,
        not isolated attributes
        
        suitable software is available
      - Disadvantages
        
        assumption of high involvement
        
        attribute importance depends on imposed range of attribute levels
        
        only small numbers of attributes can be included
      - Extension: choice-based conjoint
        
        choice is cognitively less demanding
        
        accomodates no-choice option
        
        more realistic: consumers make choices in real life
      - 3 context effects
        
        attraction: to be similar but SUPERIOR increases choice shar
        
        compromise: to be the MIDDLE option increases choice share
        
        dissimilarity: to be DISSIMILAR increases choice share
      - Example: Rooderkerk et al, 2011 - digital cameras
  - - - manipulate conditions within groups
    - - based on secondary data over time
      - longitudinal objective data collected by companies
      - 2 levels of analysis:
        
        aggregate: performance
        brand sales, store sales
        
        individual: behaviour
        brand choices, purchase timing
      - Example: Van Heerde et al, 2000:
        post & prepromotional dips (падения)
    - - monitor reactions of participants
    - - based on published empirical work
      - combine & contrast results from published empirical studies to identify systematic patters
      - explain differences in empirical findings across studies by underlying factors
      - Steps:
        
        choose moderators
        
        analyze data
        
        choose summary statistic
        (needs to be comparable across studies)
        
        present findings
        
        locate studies
        
        select relation
      - Cons
        
        differences across studies
        
        inclusion of bad studies
        
        lot of work
        
        publication bias
      - Pros
        
        high level of generalizability
        (combination of many studies)
        
        highly cited
        
        extract "structure" from large number of articles
        (not overwhelming)