Please enable JavaScript.

Coggle requires JavaScript to display documents.

Traditional Tools, Unstructured data - Coggle Diagram

- - - - Studies aimed at understanding complex, temporal processes
        (Why and How)
      - Exploratory studies. Discovering relevant constructs during theory building stage
      - For studies where the experiences of participants and context of actions are critical
    - - Group
      - Organization
      - Individual
    - - Single case
        
        Unique of extreme situations
        
        Revelatory
        
        Theory generation
        
        Critical or contrary case for testing a well-formulated theory
      - Multiple case
        
        Establishing generalizability or inferences
        
        Developing richer and more nuanced interpretations of a phenomenom
        
        Theory testing
- - - - Types
        
        Telephone
        
        Face to face
        
        Focus group:
        6 to 10 people
        
        Interviewer as facilitator
        
        Deeper examination of complex issues:
        Respondents build off one another
        
        Can be dominated by
        dominant personalities
        Facilitator needs to ensure
        everyone is involved
        
        Small size
        better for exploratory,
        not explanatory research
      - Pros
        
        Room for observations
        and comments
        
        More personal
        
        Allows for clarification of questions
        and follow-up questions
      - Cons
        
        Time consuming and
        resource-intensive
        
        Require special
        interviewing skills
        
        Interviewer needs to not
        bias observations
      - Role of Interviewer
        
        Motivate respondents
        
        Clarify and confusion or concerns
        
        Locate and enlist the
        cooperation of respondents
        
        Observe quality of response
        
        Prepare for the interview
        The data is only as good as the interviewer
    - - Group-Administered:
        Completed in a common place
        at a common time
        
        Pros
        
        High response rate
        
        Clarification if needed
      - Self-Administered:
        Mailed to large numbers of people
        
        Pros
        
        Inexpensive
        
        Unobtrusive
        
        Cons
        
        Low response rate
        
        Delayed responses
        
        Not useful for detailed responses
        or those requiring clarification
      - Web based
        
        Pros
        
        Inexpensive
        
        Modifiable
        
        Easy to complete
        
        Cons
        
        Can be compromised
        by multiple submissions
        
        Sampling bias
        i.e. must have computer
        and web access
      - Response formats
        
        Ordinal, two or more ORDERED options:
        i.e. highest level of education: HS, College, Graduate, etc.
        
        Interval-level
        i.e. 5-point or 7-point Likert Scale
        
        Nominal, two or more UNORDERED options:
        i.e. Industry of employment: education, healthcare, etc.
        
        Continuous, meaningful zero
        i.e. Age, tenure, etc.
        
        Dichotomous
        i.e. yes/no or true/false
      - Question content and wording
        
        Should not be worded in negative manner - too confusing
        
        Should not be ambiguous
        
        Must be clear and understandable
        
        Should not have biased or value-laden words
        
        Do not asked double-barreled questions
        Can have multiple answers for one respondent
        
        Do not ask questions that are too general
        
        Do not ask questions that are too detailed either
        
        Do not be presumptuous
        
        Do not ask imaginary questions
        
        Make sure your respondents have the info
        needed to answer a question
      - Sequencing
        
        Do not start with open ended questions
        
        Be chronological if asking historical questions
        
        Start with easy, non-threatening questions
        
        One topic at a time
        
        Use contingency questions if necessary
  - - - Non-Response
        
        Strategies for reducing
        
        Proper training of interviewer for recruitment
        
        Endorsement from senior leadership
        If institutional
        
        Incentives
        
        Respondent friendly questionnaires:
        Short, clear, non-offensive, easy to read and respond
        
        Non-monetary incentives
        i.e. for businesses
        
        Ensure content is relevant
        
        Confidentiality and privacy
        
        Provide advanced notification
        
        Follow-up requests
      - Sampling
        
        Web based:
        Prioritize students and younger people
        Exclude elderly and poor
        
        Questionnaires exclude young
        and illiterate
        
        Telephone:
        Prioritizes those with listed numbers
        and home during the day
        
        Surveying wrong population
      - Social Desirability
        
        People inherently do not want to look
        bad or make others look bad
        Difficult to address in questionnaires
        Can be addressed with an astute interviewer
      - Recall:
        Difficult to remember what happened in the past
      - Common Method
- - - - Prone to sampling bias
      - Not generalizable to population
      - Does not allow for estimation of
        sampling errors
    - - Expert Sampling
        
        Choosing individuals who understand
        the subject matter
        
        Not generalizable, but kind of the point
      - Snowball sampling
        
        Identify respondents meeting
        inclusion criteria
        
        Further respondents recommended by
        first respondents
        
        Useful for hard to reach populations
      - Quota sampling
        
        Population divided into strata
        
        Observations from each subgroup
        chosen non-randomly
        
        Non-proportional - Do not need proportional
        representation
        
        Proportional - proportion of respondents from each strata
        matches the proportions seen in population
        
        Still not representative of population given
        non-random nature of selection
      - Convenience Sampling
        
        Not necessarily representative of
        population
        
        Sample drawn from part of population close at hand,
        readily available or convenient
        
        Useful for pilot testing
  - - - Every unit has a non-zero
        probability of selection
      - Sampling procedure involves
        random selection
    - - Single Stage
        
        Cluster
        
        Used when population dispersed
        i.e. geographically
        Group into clusters and then measure ALL
        units within the cluster
        
        Depending on variations between clusters
        this may not be overly generalizable
        
        Systematic
        
        Prevents over representation of groups at
        either spectrum of ordering (i.e. large/small)
        
        Population is ordered based on present criteria
        Select random start and then select every kth element
        
        Simple Random
        
        Every unit has same
        probability of selection
        NO BIAS
        
        Simplicity as strength
        
        Unbiased = most
        generalizable
        
        Matched-Pairs
        
        Categorize units into two categories
        i.e. high and low profitability
        Randomly select unit from one subgroup and match with randomly selected unit from other subgroup
        
        Good for understanding differences between different
        subgroups within population
        
        Stratified
        
        Non-proportional
        Select same number of units from
        each strata regardless of overall representation
        
        Proportional
        Select more units from strata with higher percentage
        of total population in order to be more generalizable
        
        Begin by grouping into strata
        (i.e. small, medium large)
        Select x number of units from each strata
      - Multi-Stage
        
        Combing two or more Single-Stage
        sampling techniques
- - - - Inferior to Experimental design
        in internal validity due to presence
        of variety due to selection related threats
        
        Selection-Maturation Threat
        Two groups mature at different rates
        
        Selection-History Threat
        Two groups are impacted differently
        by extraneous or historical events
        
        Selection-Regression Threat
        Two groups regress toward the mean
        between pre and post test at different rates
        
        Selection-Instrumentation Threat
        Two groups respond differently
        to the measurement
        
        Selection-Testing Threat
        Both groups respond
        differently to pretest
        
        Selection-Mortality Threat
        Different drop out rates
        between two groups
      - Types
        
        Nonequivalent groups design (NEGD)
        Quasi-equivalent of pretest-posttest control group design
        
        Nonequivalent switched replication design
        Quasi-experimental version of switched replication design
        
        Regression-discontinuity (RD) design
        Assignment to treatment group based on cutoff score on a preprogram measure
        i.e. Sickest patients or poorest students assigned to treatment group
        
        Proxy pretest design
        Similar to NEGD
        Pretest score is collected after the treatment is administered
        
        Separate pretest-posttest samples design
        If not possible to collect pretest and posttest data from same subjects
        Poor internal validity
        
        Nonequivalent dependent variable (NEDV) design
        Single-group pre-post quasiexperimental design with two outcome measures
    - - Two group
        One treatment group
        One control group
        Single independent variable
        
        Pretest-posttest control group design
        Pretest - Treatment -Posttest
        or Pretest - No treatment - Posttest
        
        Statistical analysis: ANOVA
        
        Addresses maturation, testing,
        selection and regression threats
        
        Still subject to mortality threat
        
        Posttest only control group design
        Treatment - Posttest
        or No treatment -Posttest
        
        Statistical analysis: ANOVA
        
        Controls for maturation, testing, regression
        selection and pretest-posttest interction
        
        Mortality threat sill exists
        
        Covariance designs
        Similar to Pretest-Postest, but Pretest measures
        effects of covariance and not the dependent variable
        Covariance test - Treatment -Posttest
        or Covariance test - No treatment - Posttest
        
        Statistical analysis: ANCOVA
      - Factorial designs
        Two or more independent variables
        Four or higher group designs
        
        Main effect: dependent variable shows a significant
        difference between multiple levels of one factor
        at ALL LEVELS of other factors
        
        Interaction effect: The effect of differences in one factor
        depends upon the level of the second factor
        
        Not meaningful to interpret main effects
        if interaction effects are significant
      - Hybrid experimental designs:
        Combine features of other designs
        
        Solomon four-group design
        
        Two treatment and two control groups
        One treatment group and one control group are pretested
        Looking for pretest bias
        
        Switched replication design
        
        Two group design
        Two phases
        Three waves of measurement
        
        Treatment group in first phase
        becomes control group in second
        phase and vice versa
        
        Randomized block design
        
        Population grouped into subgroups (blocks)
        i.e. University students and Full time workers
        Then proceed with Pretest-Posttest or Posttest only
        for each subgroup
        
        Reduces variance in data due to attributable
        differences between the two blocks
  - - - STRONG with field experiments
      - WEAK with Laboratory experiments
    - - Maturation threat
        Observed effects due to natural
        maturation of subjects
      - Testing threat:
        Post test responses conditioned
        by pretest responses
      - History threat:
        Observed effects due to historical
        event rather than treatment
      - Instrumentation threat:
        Change in pretest and post test
        due to change in administration of
        test and not due to treatment
      - Mortality threat:
        Subjects drop out of study
      - Regression threat:
        Statistical tendency to regress
        to the mean regardless of treatment
- - - - Parsing
      - Speech Recognition
      - Generating
- - - - Unstructured
        No clear format in storage (80%)
        
        Machine
        
        Human generated
      - Semi-Structured
      - Structured
        Already stored (20%)
        
        Humans
        Data entered by humans into computer
        
        Machines