Please enable JavaScript.

Coggle requires JavaScript to display documents.

stattas - Coggle Diagram

- - - - Checks comparability of groups
    - - Both groups increase at posttest
      - Experimental group lower than control group at pretest -> both equal at posttest
      - Crossover effect
    - - Selection biases cna be reduced by matching the groups on important variables
        
        Often impossible to match on all variables
        
        In some studies, matching can cause regression to the mean
  - - - Establishes baseline functioning
    - - ITSD with Nonequivalent Comparison Group
        
        Exp group and Cont group receive multiple pretests to establish baseline behaviour
        
        Exp group interruption
        
        Both groups multiple posttest
        
        Introduces selection/interaction effective
      - ITSD with Switching Replications
        
        Exp groups 1 and 2 receive multiple pretests
        
        Exp group 1 receives interruption first then exp group 2
- - - - Deliberate comparison of participants based on these characteristics
  - - - Interviews
      - Focus Groups
    - - Grounded Theory
      - Thematic Analysis
      - Critical Discourse Analysis
      - Interpretative Phenomenological Analysis
  - - - High External Validity
        
        Can Supplement Experimental Data
    - - When is r unsuitable?
        
        Non-linear relationships
        
        Restriction of range
    - - Directionality Problem
      - Third Variable Problem
        
        Spurious Correlations
      - Statistical Control
        
        Partial Correlation
    - - Simple regression
      - Multiple regression
  - - - Disguised
        
        Ethical When:
        
        Participants remain Anonymous
        
        Setting with No Expectation Of Privacy
      - Undisguised
        
        Participant Reactivity
    - - May be information only accessible by an active participant or group member
      - No informed consent
      - Deception used
      - Researcher becomes Less Objective
      - Coding
        
        Clearly defined set of behaviours
        
        Guides different observers to code in the same way
        
        Interrater reliability
    - - Quantitative Data Collection > Qualitative
    - - Small-n
      - Lack internal and external validity
- - - - Trends
      - Latency
      - Level of change
    - - Most studies are large-n
      - Not interested in individual functioning
    - - Target behaviour is operationally defined
      - Baseline measure of target behaviour gathered
      - Treatment introduced which tries to alter the target behaviour
    - - Phase A
        
        Baseline
        
        Multiple Baseline Designs
        
        Across subjects
        
        Baseline measured in several participants at once
        
        Then one subject experiences treatment first, then another, then another etc
        
        Participants must be unable to influence each other
        
        Across settings
        
        Subject's behaviour measured in several settings
        
        Treatment established in one setting e.g. home, then school etc
        
        Settings must be independent
        
        Across behaviours
        
        Baseline measured for several related yet independent behaviours
        
        Treatment targets one behaviour, then another, then another etc
      - Phase B
        
        Treatment
    - - Increase internal validity
      - Not always ethical to withdraw treatment
    - - Baseline behaviour stability
      - Ensuring one variable changes at a time
      - Length of phases
        
        Steady state
        
        Not always feasible or ethical
        
        Longer studies risk of history
    - - High Internal Validity
      - Integration Into Practice
    - - Subjective Data Analysis
      - Low External Validity
- - - - Can be costly if researchers travel to the interviews or pay for participants to travel to the interviews
      - Telephone/interviews cheaper (if viable
      - Can be costly if hard copies need printing and, for postal questionnaires, stamps/envelopes needed
      - Online questionnaires are cheaper (if viable)
    - - Time consuming as time is spent with each individual participant (reading them questions/recording their responses)
      - Potentially time-consuming if face-to-face and administered individually
      - Postal/online questionnaires can be time consuming as researchers wait for responses
      - Deadlines or reminders may help
    - - Non-response levels high if recruitment is opportunistic but lower if interviews pre-arranged
      - Non-response levels high for postal/online questionnaires
      - May be reduced with pre-paid return envelopes or incentives
    - - Less likely for missing data
        
        Researcher asks questions, ensuring participants do not skip over any
        
        Participants can ask for clarification about unclear questions they may otherwise not answer
      - Reasons for non-response can be determined (q. Too personal etc)
      - Data more likely to be missing if none of the above applies
      - Can be reduced if online questionnaires stop skipping questions + if participants are allowed to contact the researchers if they have queries about questions
    - - Level of perceived anonymity
      - Researchers physical attributes (gender, race, attractiveness) can influence participant responses
      - Participants can provide socially desirable answers
      - Researcher effects unlikely if human contact is limited
      - Socially desirable answers are less likely
  - - - Researcher reads out the questions and writes down/records participants' answers'
      - Participants read the question themselves and write down their answer in the space provided
      - Strengths
        
        Useful when exploring a new research topic
        
        Even if you are an expert in the topic, participants can provide unexpected and insightful responses
        
        Do not provide participants with selection of answers, so are less likely to bias responses (as participants may feel they have to choose a provided answer even if none fully apply
        
        Participants' knowledge of a topic can be gauged
        
        Those with little understanding of the question cannot choose random answers or pretend they know more than they do
      - Weaknesses
        
        Participants often skip them as writing is time-consuming.
        
        Open questions best used in interviews with verbal responses
        
        Participants can provide irrelevant/vague/rambling answers
        
        They are time consuming to analyse as you must read through/listen to each answer and pick out relevant information
    - - Researcher reads out questions/response options/records answers
      - Participants reads the questions/response options/records answer
      - Types of Closed Questions
        
        Category
        
        List
        
        Rank
        
        Likert-like
        
        Verbal Labels
        
        Numerical Labels
      - Strengths
        
        Can be answered quickly (participants like the items -> reduces missing data)
        
        Having options available may make ambiguous questions easier to understand
        
        They eliminate vague/irrelevant/rambling answers
        
        They are reasonably quick to analyse (relative to open questions)
      - Weaknesses
        
        Participants may want to provide an answer that is not an option
        
        Participants can choose random answers, even if they do not understand questions
- - - - Clear Definition
      - Inclusion / Exclusion Criteria Set
    - - Questionnaires
      - Psychometrics Tests
      - Laboratory Tasks
  - - - Test-Retest Coefficient
        
        Ideally above .80 but no lower than .70
      - Does a test provide a consistent measure of a stable construct if used more than once?
      - Considerations
        
        Unstable Constructs (Mood)
        
        Carryover Effects (Practice Effect)
    - - If individual test items measure one construct, do they produce results consistent with one another?
      - Cronbach's Alpha :wolf:
        
        Values above .70 are acceptable
        
        Values below .70 suggest participants' responses to 1+ items differed in their responses on other items
      - Interitem Correlation
    - - Do different researchers, using the same test to mesure the same construct in a person, produce consistent results?
      - Cohen's Kappa :turtle:
        
        Categorical Data
        
        Assumptions
        
        Judgements are Discrete
        
        Judgements based on Paired Observations (i.e. Same Target)
        
        Judgements are based on the Same Number of Categories
        
        Judges are Independent
        
        The Same Judges are used for all observations
        
        Ranges 0 - 1
        
        0 = Chance Agreement
        
        1 = Complete Agreement
        
        <0 = More Disagreement than Chance
      - Intraclass Correlations (ICC)
        
        Continuous Data
  - - - Does a test seem to measure a construct
      - Sometimes warranted depending on purpose
      - Naive + Expert Agreement on what the test measures = Good Face Validity
    - - Does a test measure every element of a construct?
      - Unidimensional or Multidimensional
      - Lawshe, 1975
        
        Suggest giving tests to experts -> asking them if each item is either Essential, Useful, or Not Necessary. If >50% class an item as essential, it has content validity
    - - Predictive Validity
        
        Performance on a new test predicts future behaviour or outcome
        
        Criterion assumed to be valid
        
        Established when the new test successfully predicts an associated future behaviour
      - Concurrent Validity
        
        Performance on a new test predicts performance on an outcome related to the new test
        
        Criterion is assumed to be valid
        
        Data are collected at same or nearly the same time
      - Do scores on a test correlate with other measures/criterions you would reasonably expect them to correlate with?
    - - Does the test actually measure the construct?
      - Relies on measurement of other validities
      - Four Common Methods:
        
        Differential Group Studies
        
        Assess the construct validity by examining the presence of a construct
        
        True Positive (Group 1) and True Negative (Group 2)
        
        Test confirms validity if it confirms only Group 1 possesses the construct
        
        Intervention Studies
        
        Test should confirm presence of the construct in people known to have it
        
        A treatment that changes the construct is provided. The test should capture the change
        
        Tests of Convergent Validity
        
        Scores on a test measuring a construct should positively correlate with scores on tests measuring theoretically the same/similar construct
        
        Tests of Divergent Validity
        
        Scores on a test measuring a construct shoudl fail to correlate or negatively correlate on tests of theoretically dissimilar constructs
- - - - Threat in BW-S studies
    - - Mitigates history effect
  - - - Make EVs into CVs
- - - - Missing Data
        
        Participant did not provide data
        
        There was a problem with the study
        
        If online, program may have broken
        
        Researchers lost/accidentally deleted data
        
        Pituch & Stevens (2016)
        
        Excluding Data
        
        Listwise Deletion
        
        Excludes participants with missing data from all analysis
        
        Casewise Deletion
        
        Excludes participants from analysis involving the variable with the missing data only
        
        Imputation
        
        Replacing Missing Data with the Mean
      - Incorrect Scores
  - - - Original researchers may have fabricated data
      - Original researchers may have deliberately analysed data incorrectly to obtain interesting findings
      - Fluke / One-off findings (Type 1 Errors)
    - - Hoekstra et al. (2012)
        
        Few researchers actually check for assumption of normality of their data
  - - - Komolgorov-Smirnov (KS)
      - Shapiro-Wilk (SW)
      - Vulnerable to sample-size
    - - Outliers
        
        Boxplots (Identify)
        
        Consider Reason for Outlier/s
        
        Participant did not follow instructions
        
        Researcher did something different
        
        Technical Issue
        
        Participant did not pay attention
        
        Participant responded mischievously
        
        Participant was Just That Good
        
        Windsorising
        
        Substitutes an outlier with the highest value that isnt an outlier
        
        Can always describe and justify windsorising
        
        Better option if trying to keep sample size
        
        Trimming
        
        Deletes highest and lowest scores (5% - 10%)
        
        Worst option if you dont want to reduce sample size
        
        In RT studies, responses 2.5 SDs above the mean are trimmed
      - Skew
        
        Histogram
        
        Skew isn't alwasy problematic -- consider the construct being measured
        
        Transformation
        
        Tests of relationships between two or more variables
        
        Only transform problematic variables
        
        Tests of differences between scores or changes in scores over time
        
        Transform All DVs
        
        Compare your data's distribution and decide manually which distribution the data most closely resembles
        
        Use Non-Parametric Test