Pre-lab notes: mood data (MOOD) and split reliability

JASP pre-lab notes: mood dataset (MOOD) and reliability (SPLIT)

  • Purpose and mindset

    • JASP is a statistical program designed for psychology; aims to show where buttons are and explain what they mean, helping you learn by doing.
    • Statistics are a tool to investigate what’s going on in people's heads and test ideas beyond intuition, which can be misleading.
    • If you have a preferred learning style, consider using the recommended textbook for clearer explanations.

  • Datasets used in the session

    • MOOD: used for descriptive statistics and correlation analysis.
    • SPLIT: used to learn about reliability of a measure via Cronbach Alpha.
    • Two data types appear in MOOD: a categorical variable (gender, labeled as sex) and several continuous variables (happy, pos, n).

  • Mood file: variables and interpretation

    • Gender variable: sex (binary male/female); categorical; means are not meaningful. Use frequencies in reports (e.g., gender counts).
    • Continuous variables:
    • happy: life happiness (general happiness)
    • pos: positive mood (mood in the moment)
    • n: neuroticism (personality trait, not clinical; tendency to negative emotions and emotional volatility)
    • Neuroticism notes:
    • Part of BFI; often 12 items (6 positive, 6 negative).
    • Negative items should be reverse-scored and aggregated into a single neuroticism score.
    • Data structure:
    • Each row = a participant; each column = a variable (e.g., participant 1 has a neuroticism score of 37; happiness and mood scores follow similarly).
    • Visual intuition:
    • Mood tends to vary; may show greater variance than trait-like happiness.

  • Quick pre-lab exercise: hypothesis generation

    • Think of possible relationships between mood, happiness, and neuroticism; relationships may be positive or negative.
    • Discuss the idea that relationships could be linear and that there is a spectrum of individuals.

  • Descriptive statistics in MOOD (descriptives)

    • In JASP, open MOOD, select variables, and add them to the Descriptives analysis box; you can drag individually or select all at once.
    • Output on the right shows descriptors; means and standard deviations (SD) for continuous variables are highlighted in green.
    • Gender is treated as categorical; report frequencies instead of means.
    • Interpreting the descriptive stats:
    • Mean = average score across participants.
    • SD = how much scores vary around the mean.
    • Practical note: raw average scores don’t imply high/low meaning without population norms; avoid claiming that “neuroticism levels are very high/low” without norms.
    • Reporting tip: present means and SDs for happy, pos, and n; report gender frequencies.

  • Understanding and interpreting scatter plots

    • Scatter plots visualize relationships, not tests; eyeballing is useful but unreliable.
    • Example concepts explained in the video:
    • Temperature vs drinks sold: positive association (as temperature rises, drink sales rise).
    • Temperature vs coffee sales: negative association (as temperature rises, coffee sales fall).
    • Mood data visuals:
    • happy vs pos: positive association (high life happiness tends to coincide with higher positive mood).
    • n (neuroticism) vs happy: negative association (higher life satisfaction tends to relate to lower neuroticism).
    • n vs pos: downward tendency (more neuroticism tends to be associated with lower positive mood, though with more variance).
    • Important caveat: scatter plots suggest patterns but do not establish significance or causation.

  • Pearson correlation: purpose and interpretation

    • Purpose: statistically test for a linear relationship between two continuous variables.
    • Notation and range:
    • Correlation coefficient r ∈ [-1, 1].
    • Positive r indicates a positive relationship; negative r indicates a negative relationship.
    • Larger |r| indicates a stronger relationship.
    • Mathematical expression (conceptual):
      r = \frac{\mathrm{cov}(X,Y)}{\sigmaX \sigmaY}
    • Significance: report the p-value to indicate whether the observed correlation is statistically significant (often p < 0.05 is used as a threshold).
    • Two-tailed test: the test assesses the possibility of both positive and negative associations.
    • Reading the JASP output: a two-by-two matrix with the reported r value, the sample size n, and the p-value for the pair of variables.
    • Example from MOOD data:
    • Life satisfaction (happy) and neuroticism (n): r = -0.413 \quad p < 0.05
    • Interpretation: a significant negative relationship; higher life satisfaction tends to be associated with lower neuroticism; n indicates the sample size used for this correlation.
    • Effect size interpretation (Cohen’s guidelines, context-dependent):
    • Roughly, small ~ |r| ≈ 0.10, medium ~ |r| ≈ 0.30, large ~ |r| ≈ 0.50.
    • The example r = -0.413 is described as a medium effect size in the session.
    • Practical takeaway: correlations give you an effect size (r) and significance (p); do not rely on p-values alone; consider the magnitude of the effect and theoretical justification.

  • Spurious correlations: why significance isn’t enough

    • It’s possible to obtain statistically significant correlations that are meaningless if there is no theoretical basis.
    • Example described:
    • Per capita mozzarella cheese consumption and number of lawyers in Hawaii showed a very high correlation (r ≈ 0.97) with no theoretical link.
    • Takeaway: avoid “fishing” for significant correlations; rely on theory or prior data to justify the expected relationship.
    • In psychology: test hypotheses grounded in theory or prior literature (e.g., personality and self-determination theory) to avoid spurious findings.

  • Reliability and Cronbach Alpha (SPLIT dataset)

    • Reliability in this context means internal consistency: do items that are supposed to measure the same construct yield similarly shaped responses?
    • Internal consistency is about the measurement instrument, not about the participants.
    • Why reliability matters:
    • Many scales are multi-item; we want those items to tap into the same underlying construct (e.g., extraversion) so that aggregation makes sense.
    • If some items diverge, it may indicate multiple constructs are being measured or items don’t fit the construct.
    • SPLIT data structure:
    • Data include items related to different splitting domains: self (s), family (f), others (o).
    • Example items show how items map to subscales (e.g., family-splitting items 2 and 4). Sorting helps identify which items belong to which subscale.
    • Cronbach Alpha (concept): an index of internal consistency across items in a subscale.
    • How to run Cronbach Alpha in JASP (SPLIT file):
    • Open the SPLIT file; switch to Reliability (blue top bar).
    • Choose classical reliability (Cronbach Alpha).
    • Alphabetize variables (A to Z) to find subscale items easily.
    • Select items belonging to the subscale (e.g., all family-splitting items starting with 'f').
    • Ensure Cronbach Alpha is selected (not McDonald’s omega or other options).
    • Example result:
    • Family-splitting subscale Cronbach Alpha ≈ 0.83.
    • Interpreting Cronbach Alpha (George & Mallory guidelines, used in the session):
    • > 0.9: excellent
    • 0.8 – 0.9: good
    • < 0.5: unacceptable
    • (Note: other categories exist in literature; these guidelines are simplified for teaching purposes.)
    • Interpretation for the SPLIT example:
    • α ≈ 0.83 for family-splitting items → good reliability; the subscale is likely tapping into the same underlying construct of family-splitting.
    • Additional reporting notes:
    • Reliability is typically reported in the method section and pertains to the measures, not the participants.

  • How to report findings and integrate theory

    • When reporting, combine descriptive stats, correlations, and reliability with a theoretical frame (e.g., literature on personality and mood).
    • Typical reporting structure:
    • Descriptives: means and SDs for continuous variables; gender frequencies.
    • Correlations: report r, p-values, and n; interpret direction and magnitude; discuss significance in context.
    • Reliability: report Cronbach Alpha for subscales; interpret using guidelines.
    • Emphasize that correlations are evidence for relationships but do not imply causation; use theory to justify expected directions.
    • Learn to distinguish between aggregated mood scores (MOOD) and item-level data (SPLIT) when performing analyses.

  • Practical tips for the lab session

    • Practice by pausing the video after performing each step in JASP to match the instructor’s actions.
    • Write down specific problems and questions to bring to tutors in the next lab.
    • Keep in mind the difference between descriptive statistics, visualization, correlation, and reliability analyses; each serves a different purpose.

  • Quick recap of key concepts and terms

    • Descriptives: mean, standard deviation; reporting for continuous variables; frequencies for categorical variables.
    • Scatter plots: visualization of relationships; not a substitute for statistical tests.
    • Pearson correlation (r): strength and direction of a linear relationship; range [-1, 1].
    • p-value: probability of observing the data (or more extreme) under the null hypothesis; two-tailed test used in this session.
    • Cronbach Alpha (α): measure of internal consistency reliability for a set of items; higher values indicate more coherent measurement of a single construct.
    • Internal consistency vs construct validity: reliability is about measurement consistency, while validity concerns whether the measure captures the intended construct.

  • Final notes

    • The pre-lab videos are designed to prepare you for the lab by outlining how to use JASP and how to interpret outputs.
    • If you had trouble, identify concrete questions and discuss them with tutors or in the next lab.

  • References to the workflow mentioned in the video (for quick recall)

    • Load MOOD data and run Descriptives; add scatter plots; interpret means and SDs.
    • Run Pearson correlation for selected pairs (e.g., happy-n; happy-n, etc.); report r, p, and n; classify effect size.
    • Load SPLIT data; sort items by subscale; run Cronbach Alpha for a chosen subscale; interpret α against guidelines.

  • Notation and formulas used

    • Pearson correlation coefficient:
      r = \frac{\mathrm{cov}(X,Y)}{\sigmaX \sigmaY}
    • Cronbach Alpha (conceptual formula):
      \alpha = \frac{N \bar{c}}{\bar{v} + (N-1) \bar{c}}
    • Relationship interpretation: from the MOOD example, the reported correlation was
      r = -0.413, \quad p < 0.05,
      indicating a significant negative relationship with a medium effect size per Cohen’s guidelines.