Theory of Mind × Empathy — Cross-instrument analysis · GTEMO Experiment

Author

Eric Guerci

Published

March 22, 2026

1 Objective

This page tests whether self-reported empathy (IRI) is associated with film-based Theory of Mind performance (MASC). The two instruments capture related but distinct constructs:

  • MASC — implicit, behavioural ToM measured via naturalistic film clips (17 affective items + 28 cognitive items = 45 total)
  • IRI — explicit, self-reported empathic disposition across four subscales (Perspective Taking, Empathic Concern, Fantasy, Personal Distress)

Two complementary analyses are reported:

  1. Level A — Spearman correlations: targeted pairings to check whether matching pairs (affective ToM ↔︎ affective empathy; cognitive ToM ↔︎ cognitive empathy) are stronger than crossing ones (discriminant validity)
  2. Level B — Binomial GLMs: IRI subscales entered simultaneously as predictors of MASC accuracy; a quasi-binomial robustness check corrects for potential overdispersion

2 Level A — Spearman correlations

Show code 
tab_spearman |>
  gt() |>
  cols_label(
    MASC_dim = "MASC dimension",
    Pair     = "Pair",
    rho      = "\u03c1",
    p_fmt    = "p",
    sig      = "Sig."
  ) |>
  tab_header(
    title    = "Spearman correlations: MASC \u00d7 IRI",
    subtitle = paste0("N = ", n_mi,
                      " complete cases. Exact = FALSE (ties present).")
  ) |>
  tab_style(style = cell_text(weight = "bold"),
            locations = cells_column_labels()) |>
  tab_style(style = cell_text(weight = "bold"),
            locations = cells_body(columns = sig, rows = sig != "ns")) |>
  tab_footnote("* p < .05  ** p < .01  *** p < .001  ns = not significant")
Spearman correlations: MASC × IRI
N = 122 complete cases. Exact = FALSE (ties present).
MASC dimension Pair ρ p Sig.
Affective ToM Affective ToM × Empathic Concern -0.002 0.982 ns
Affective ToM Affective ToM × Personal Distress -0.047 0.605 ns
Affective ToM Affective ToM × Perspective Taking -0.148 0.103 ns
Cognitive ToM Cognitive ToM × Perspective Taking 0.041 0.652 ns
Cognitive ToM Cognitive ToM × Empathic Concern -0.030 0.742 ns
Cognitive ToM Cognitive ToM × Fantasy -0.045 0.622 ns
Cognitive ToM Cognitive ToM × Personal Distress 0.025 0.787 ns
* p < .05 ** p < .01 *** p < .001 ns = not significant
Note

Matching vs crossing hypothesis. Affective ToM (emotion inference from film clips) is theorised to align more strongly with affective empathy (Empathic Concern, Personal Distress). Cognitive ToM (belief/intention inference) should align more with cognitive empathy (Perspective Taking). Pairs that cross the affective/cognitive boundary serve as a discriminant validity check — weaker or non-significant ρ there supports construct differentiation.

2.1 Correlation heatmap

Show code 
p_cor_heat
Figure 1: Spearman ρ between the two MASC dimensions (rows) and the four IRI subscales (columns). Red = positive association, blue = negative. Significance stars: * p < .05 ** p < .01 *** p < .001.

3 Level B — Binomial GLMs

IRI subscales entered simultaneously as predictors of MASC accuracy. The response is modelled as a binomial count of correct answers (17 affective items; 28 cognitive items, total = 45). Coefficients are on the log-odds scale; the forest plot shows exponentiated odds ratios (OR) with 95% Wald CIs.

Show code 
tab_glm |>
  select(Outcome, Predictor, beta, SE, OR, OR_lo, OR_hi, stat, p_fmt, sig) |>
  gt() |>
  tab_header(
    title    = "Binomial GLM: IRI subscales predicting MASC accuracy",
    subtitle = "Family: binomial (logit link). Wald 95% CI."
  ) |>
  cols_label(beta = "\u03b2", SE = "SE", OR = "OR",
             OR_lo = "95% CI lo", OR_hi = "95% CI hi",
             stat = "z", p_fmt = "p", sig = "Sig.") |>
  tab_row_group(label = "Outcome: Cognitive ToM (28 items)",
                rows = Outcome == "Cognitive ToM") |>
  tab_row_group(label = "Outcome: Affective ToM (17 items)",
                rows = Outcome == "Affective ToM") |>
  tab_style(style = cell_text(weight = "bold"),
            locations = cells_column_labels()) |>
  tab_style(style = cell_text(weight = "bold"),
            locations = cells_body(columns = sig, rows = sig != "")) |>
  tab_style(style = cell_text(weight = "bold", color = "#2d7a3a"),
            locations = cells_row_groups()) |>
  tab_footnote("\u03b2 = log-odds coefficient. OR = exp(\u03b2). Wald 95% CI. * p < .05  ** p < .01  *** p < .001.")
Binomial GLM: IRI subscales predicting MASC accuracy
Family: binomial (logit link). Wald 95% CI.
Outcome Predictor β SE OR 95% CI lo 95% CI hi z p Sig.
Outcome: Affective ToM (17 items)
Affective ToM Empathic Concern 0.0061 0.0124 1.006 0.982 1.031 0.488 0.626
Affective ToM Perspective Taking -0.0160 0.0116 0.984 0.962 1.007 -1.380 0.167
Affective ToM Fantasy 0.0001 0.0093 1.000 0.982 1.018 0.015 0.988
Affective ToM Personal Distress -0.0041 0.0094 0.996 0.978 1.014 -0.444 0.657
Outcome: Cognitive ToM (28 items)
Cognitive ToM Empathic Concern 0.0001 0.0096 1.000 0.981 1.019 0.014 0.989
Cognitive ToM Perspective Taking 0.0076 0.0088 1.008 0.990 1.025 0.867 0.386
Cognitive ToM Fantasy -0.0081 0.0071 0.992 0.978 1.006 -1.133 0.257
Cognitive ToM Personal Distress 0.0062 0.0072 1.006 0.992 1.020 0.860 0.39
β = log-odds coefficient. OR = exp(β). Wald 95% CI. * p < .05 ** p < .01 *** p < .001.
Note

Overdispersion check. A binomial GLM assumes variance = μ(1−μ)/n; real data often show extra-binomial variation. The dispersion parameter φ is estimated by the quasi-binomial fit: φ(affective) = 0.615, φ(cognitive) = 0.817. φ ≈ 1 means the binomial assumption holds; φ >> 1 means standard binomial SEs are underestimated.

Show code 
p_glm_forest
Figure 2: Forest plot: odds ratios from the binomial GLMs. Error bars = 95% Wald CI. Dashed line = OR 1 (null effect).

4 Quasi-binomial robustness check

The quasi-binomial model uses the same formula but estimates a free dispersion parameter φ, inflating standard errors by √φ. Coefficients (β) and odds ratios are identical to the binomial — only SEs and p-values change. The comparison table shows directly where overdispersion changes inference.

Show code 
tab_glm_compare |>
  select(Outcome, Predictor, beta, OR,
         SE_binom, SE_quasi, SE_ratio,
         p_binom, sig_binom, p_quasi, sig_quasi) |>
  gt() |>
  tab_header(
    title    = "Binomial vs quasi-binomial: SE and p-value comparison",
    subtitle = paste0("\u03c6 (dispersion): Affective = ", disp_aff,
                      ", Cognitive = ", disp_cog,
                      ". SE ratio \u2248 \u221a\u03c6.")
  ) |>
  cols_label(
    beta      = "\u03b2", OR = "OR",
    SE_binom  = "SE (binom)", SE_quasi = "SE (quasi)", SE_ratio = "SE ratio",
    p_binom   = "p (binom)",  sig_binom = "Sig. (binom)",
    p_quasi   = "p (quasi)",  sig_quasi = "Sig. (quasi)"
  ) |>
  tab_row_group(label = "Outcome: Cognitive ToM",
                rows = Outcome == "Cognitive ToM") |>
  tab_row_group(label = "Outcome: Affective ToM",
                rows = Outcome == "Affective ToM") |>
  tab_style(style = cell_text(weight = "bold"),
            locations = cells_column_labels()) |>
  tab_style(style = cell_text(weight = "bold", color = "#2d7a3a"),
            locations = cells_row_groups()) |>
  tab_style(
    style = cell_fill(color = "#fff3cd"),
    locations = cells_body(
      columns = c(sig_binom, sig_quasi),
      rows = sig_binom != sig_quasi
    )
  ) |>
  tab_footnote("Yellow highlight = significance changes between models. SE ratio = SE\u2098\u1d64\u1d43\u02e2\u1d35 / SE\u1d47\u1d35\u207f\u1d52\u1d50.")
Binomial vs quasi-binomial: SE and p-value comparison
φ (dispersion): Affective = 0.615, Cognitive = 0.817. SE ratio ≈ √φ.
Outcome Predictor β OR SE (binom) SE (quasi) SE ratio p (binom) Sig. (binom) p (quasi) Sig. (quasi)
Outcome: Affective ToM
Affective ToM Empathic Concern 0.0061 1.006 0.0124 0.0098 0.79 0.626 0.535
Affective ToM Perspective Taking -0.0160 0.984 0.0116 0.0091 0.78 0.167 0.081
Affective ToM Fantasy 0.0001 1.000 0.0093 0.0073 0.78 0.988 0.985
Affective ToM Personal Distress -0.0041 0.996 0.0094 0.0073 0.78 0.657 0.573
Outcome: Cognitive ToM
Cognitive ToM Empathic Concern 0.0001 1.000 0.0096 0.0087 0.91 0.989 0.988
Cognitive ToM Perspective Taking 0.0076 1.008 0.0088 0.0080 0.91 0.386 0.34
Cognitive ToM Fantasy -0.0081 0.992 0.0071 0.0065 0.92 0.257 0.213
Cognitive ToM Personal Distress 0.0062 1.006 0.0072 0.0065 0.90 0.39 0.344
Yellow highlight = significance changes between models. SE ratio = SEₘᵤᵃˢᴵ / SEᵇᴵⁿᵒᵐ.
Show code 
p_glm_forest_quasi
Figure 3: Forest plot: odds ratios from the quasi-binomial GLMs. Wider CIs reflect SE inflation by √φ. Compare with the binomial forest plot above.