Skip to content

Lazarsfeld & Simpson Lab

Compare the association between X and Y before and after holding Z constant. Edit a partial table, drag a partial bar or change the composition: the zero-order percentages are recomputed, never typed. Then classify what you see and decide whether Z is antecedent or intervening.

X = Reads news daily · Y = Votes · Z = Education

% Y = 1 by X: zero-order and within each level of Z

bars: % Y = 1 for X = 1 (blue) and X = 0 (grey); drag a partial bar to change it; d = the difference

Zero-order61.0 %X = 136.0 %X = 0d = +25.0 ppn = 400Education: no degree60.0 %X = 135.0 %X = 0d = +25.0 ppn = 200Education: degree62.0 %X = 137.0 %X = 0d = +25.0 ppn = 200
% Y = 1 by X: zero-order and within each level of Z. zero-order +25.0 pp, partials +25.0 pp and +25.0 pp

Composition of the X groups by Z

X = 1no degree 100degree 100X = 0no degree 100degree 100
Composition of the X groups by Z: X = 1: 100 / 200 in degree; X = 0: 100 / 200 in degree

The zero-order table is the weighted sum of the partials. If the X groups differ in Z and Z relates to Y, the aggregate can differ from every partial.

Partial tables (editable)

Education: no degree
Y = 1Y = 0%
X = 160.0
X = 035.0
Education: degree
Y = 1Y = 0%
X = 162.0
X = 037.0

Hand-authored illustrative counts (not an empirical finding). Identity checked on every render: the zero-order table equals the cell-wise sum of the partials. Classification uses the % differences only; the same partial pattern is 'explanation' or 'interpretation' depending on causal order, and the data cannot tell them apart.

zero-order +25.0 pp, partials +25.0 pp and +25.0 pp

Formula

d₀ = %Y(X=1) − %Y(X=0)d_z = the same within Z = zaggregate % = Σ_z w_z · %_z, w_z = share of the X group in Z = z

Replication: d_z ≈ d₀. Specification: d_z differs across z. Explanation / interpretation: d_z ≈ 0 while d₀ ≠ 0 (spurious vs mediated). Simpson: d_z reverses the sign of d₀.

Method: Partial tables; zero-order = cell-wise sum; % differences · Synthetic, seeded data. Illustrative, not an empirical finding.

Validation: Aggregate = weighted partials (1e-12) and the published kidney-stone reversal (tests/unit) · integrated