1. The method-factor deflation (critique §1): tested and rejected
VERDICT: survives re-testing
The critique's strongest move is that the third factor is common-method variance produced by several items sharing a seven-point self-placement response format, and that removing it leaves exactly the compass's two substantive axes: redistribution and authority. This is the kill-shot, and it is falsifiable. We ran two tests.
First, we re-ran the full retention pipeline after excluding the five seven-point self-placement items (the spending-services, private-vs-government insurance, guaranteed-jobs, assistance-to-Blacks, and birthright-citizenship scales). If the third factor were carried by that shared format, retention should collapse to two. It does not: parallel analysis still retains three factors on the sixteen remaining substantive items across all random seeds tested, though Velicer's MAP drops to two, indicating the third dimension is weaker but not format-dependent.
Second, and more directly, we fit a confirmatory model that specifies a dedicated method factor loading only on those five self-placement items, over and above two substantive factors. If the third dimension were method variance, isolating it into an explicit method factor would improve fit. It does not move fit at all: comparative fit index .666 with the method factor versus .663 without it, identical within noise, and the same RMSEA and SRMR to three decimals. A method factor that absorbs no misfit is not capturing a real source of shared variance. The 'F3 is common-method variance' hypothesis is therefore rejected by the data that would have confirmed it. We are grateful to the critique for naming the test; running it converts a plausible objection into positive evidence for the structure.
This does not make the third factor substantively glamorous—it remains a scope-and-direction dimension correlated with the dominant ideological factor—but it is not an artifact of question format, and the substantive structure is not simply the two-axis compass.
Figure A. The method-factor test. Left: retention is unchanged when the five seven-point self-placement items are removed (parallel analysis still returns three). Right: specifying an explicit method factor on those items does not improve confirmatory fit, rejecting the hypothesis that the third factor is common-method variance.
2. The knife-edge and the omitted number (critique §2): conceded and reported
VERDICT: conceded — revision incorporates
The critique notes that the fourth factor missed the parallel-analysis threshold by a hair (eigenvalue 1.00 against a 1.06 threshold), that such thresholds carry sampling error, and—most pointedly—that the paper reported split-half congruence for three factors but conspicuously omitted it for the fourth, which is the number that adjudicates whether a fourth factor is real-but-weak or noise.
This criticism is correct and the omission is now repaired. The fourth-factor split-half Tucker congruence is .92, which exceeds the .90 standard for factor equivalence. The fourth factor replicates. This cuts against the paper's clean 'three, not four' headline: the boundary is soft, and the data are better described as 'three robust factors plus a weak-but-replicable fourth' than as a sharp three. The revision reports the .92 congruence in the main text and reframes the dimensionality finding as a soft lower bound rather than a hard count. Notably, this concession runs against the critique's larger thesis as much as against the paper's: the data are pushing toward more structure, not the two-axis compass.
We retain the point that the four-axis confirmatory model fits poorly (CFI .70), but agree that poor fit of one specified pattern is weak evidence for exactly three dimensions, and we no longer rest weight on the precise count.
3. Factor count is battery-dependent (critique §3): conceded; the ontological claim is withdrawn
VERDICT: conceded — revision incorporates
The critique's deepest point is methodological: the number of dimensions recovered is a joint function of the items fed in, the population, and the year, so 'three dimensions' is a statement about a twenty-one-item battery in 2020, not about the structure of belief as such. The paper's own epistemic-axis null—correctly downgraded in its limitations to 'not supported by these items'—is cited as proof of the principle: change the items, change the dimensions.
This is right, and it exposes a genuine inconsistency between the paper's careful internal hedging and its confident abstract and conclusion, which escalate to 'mass political belief is organized into three dimensions.' That ontological phrasing is not licensed by the method and is withdrawn. The revised abstract and conclusion claim only that this standard ANES battery resolves into three replicable common factors in 2020, that two of the four pre-registered axes do not separate as specified, and that the substantive content is a fused ideological dimension plus a distinct authoritarian dimension. The escalation from 'this battery factors into three' to 'belief has three dimensions' was the paper's real overreach, and the critique identified it precisely.
4. Drift: composition versus content (critique §4): tested; direction survives, characterization corrected
VERDICT: mixed — direction survives, characterization corrected
The critique raises two distinct objections to the drift section. The first: a moving cross-sectional group mean cannot distinguish content drift (the label's referent migrated) from compositional change (different people adopted the label), especially as the label's prevalence changed. The second: the paper treated 'conservative' as a stable policy referent, which inverts the well-established finding that conservative identification is the more symbolic and less operational label.
On the first, we ran the decomposition the critique demanded. Two results. The share of GSS respondents self-identifying as liberal on the seven-point scale changed only modestly across the window (about +1.6 percentage points), so the prevalence explosion the critique cites—the doubling of liberal identification—is drawn from party-subset series (liberal share among Democrats) that do not correspond to this paper's estimand. More decisively, a cohort decomposition of liberals' policy positions from the pre-1990 to the post-2010 period attributes 97.6 percent of the movement to within-cohort period change and essentially none to between-cohort composition. The same birth cohorts holding the liberal label moved leftward over time. This is content drift, and it survives precisely the test designed to dissolve it.
On the second, the critique is correct and the paper was wrong. Examining operational policy positions by self-identification, only about a third of self-identified conservatives sit on the operationally liberal side of our index—but the paper's error was not about this number; it was in calling 'conservative' the 'fixed-content' pole, which conflated temporal stability with operational anchoring. The symbolic-conservative, operationally-liberal pattern documented since Free and Cantril and developed by Ellis and Stimson establishes that conservative self-identification is loosely tied to operational policy. The revision removes any claim that conservative self-identification is operationally anchored and reports only the defensible asymmetry: self-identified liberals' positions moved more over time than conservatives', a directional finding corroborated outside this paper, with no implication about which label is better tied to policy.
Finally, the critique is right that Section 2 was less rigorous than the dimensionality study and that drift is, strictly, a critique of one-dimensional self-identification rather than of the two-dimensional compass. The revision shores up Section 2 with the decomposition above and explicitly scopes the drift finding as an indictment of label-based, self-identification measurement, not of the compass's axes.
Figure B. The drift decomposition. Left: 97.6% of the change in self-identified liberals' policy positions is within-cohort period movement, not between-cohort composition. Right: operational ideology by self-identification — only a third of self-identified conservatives are operationally liberal, but the symbolic-conservative literature establishes 'conservative' as the less operationally anchored label, correcting the original paper's characterization.
5. The prescription is self-defeating (critique §5): partly accepted, partly contested
VERDICT: contested — partly accepted
The critique argues that 're-estimate the axes over time' abolishes the comparability that makes a compass useful, and that 'derive rather than assert' merely relocates the arbitrary choice upstream to the item battery. The second point is accepted: deriving structure from responses does not escape the need to choose items, and the epistemic-axis null shows that choice has teeth. The revision states this plainly and frames the recommendation as 'make the assertion explicit and testable at the item level' rather than 'escape assertion.'
The first point we contest. Re-estimating axes need not destroy comparability if successive waves are linked through measurement-invariance testing on anchor items—the standard psychometric solution to exactly this problem. One can hold a subset of marker items invariant across waves, establish partial invariance, and measure drift against that anchored scale. That is more demanding than a fixed compass but not incoherent, and it is how longitudinal measurement handles instruments whose meaning may shift. The revision replaces the unqualified 're-estimate over time' with this anchored-invariance proposal.
6. 2020 is a high-water mark (critique §6) and EFA cannot see typology (critique §7): conceded as scope limits
VERDICT: conceded — revision incorporates
Two further criticisms are accepted as bounds on generality rather than refutations. First, 2020 was a peak-sorting, high-salience year—pandemic, post-2020 racial salience—so the fusion of economic, racial, and environmental attitudes on one factor may reflect maximal momentary co-activation rather than a structural constant. A single cross-section cannot separate 'permanently fused' from 'maximally co-activated in one extraordinary year.' This caveat now appears in the headline claims, not only the limitations, and the revision explicitly frames the fusion as a 2020 observation requiring replication across calmer years (the natural next study: re-run on ANES 2012 and 2016).
Second, linear common-factor analysis can only represent dimensional structure; it is blind to typological or configural organization. A three-factor result therefore cannot establish that belief is dimensional rather than profile-based. The revision concedes that the dimensional count presupposes a dimensional ontology and reframes the contribution as 'within a dimensional framework, this battery yields three factors,' explicitly declining to adjudicate dimensional versus typological organization.
7. What the re-testing establishes
After running the critique's own decisive tests: the method-artifact deflation is rejected by direct test; the fourth-factor congruence is reported and replicates, softening the count; the ontological overreach is withdrawn; the drift direction survives a cohort decomposition while the conservative-anchoring claim is corrected as an error; the prescription is reframed via measurement invariance; and the 2020 and dimensional-ontology caveats are promoted from footnotes to headline scope conditions. The surviving contribution is narrower and more defensible than the original: in the 2020 ANES, a standard issue battery yields three replicable factors whose substantive content is a fused ideological dimension and a distinct authoritarian dimension; the popular compass's labels have demonstrably drifted and sorted; and an honest ideological map must make its item-level assertions explicit and anchor comparisons through invariance rather than asserting fixed axes. The compass's dimensionality is partly vindicated by these data, as the critique argues; its provenance and labels are not, as the paper shows. The revision claims only the second scalp.
This is a companion to Why the Popular Political Compass Fails, the study it defends.