# Pre-Registration: Divided Government and Significant Lawmaking, 1947 to Present **Status:** FROZEN 2026-07-14, before any contact with the outcome data. **Author:** IV (frankfrisbieiv-creator) **Method:** ARIA. Pre-register blind, treat missing data as ignorance not a result, halt loudly on completeness shortfalls, report nulls flat, require agreement across independent significance measures before any robust claim. This registration is written before downloading or inspecting any significance-of-legislation counts. The divided-government coding is reused from the executive-orders study, whose control table already exists; the outcome data does not. --- ## 1. Question Does divided party control of the federal government reduce the amount of significant legislation Congress enacts? The popular claim is that divided government produces gridlock and fewer major laws. Mayhew's *Divided We Govern* (1991, updated) rejected that for 1947 to 1990 and later years, and independent replications (Clinton and Lapinski 2006; Ansolabehere, Palmer, and Schneer 2018; Binder) have not overturned it. We register the expectation that the null holds, so a null is not a surprise we are dressing up, and a real gridlock effect would be the surprise. The live, un-settled version of the question is whether a gridlock effect appears specifically in the polarization era, which did not exist when Mayhew wrote. That is the secondary hypothesis, registered as underpowered. ## 2. Unit and span Unit of analysis is the Congress (two-year). Span is the 80th Congress (1947) through the last Congress covered by Mayhew's updated important-enactments list. The exact endpoint differs across sources (2014, 2018, or 2022). It is verified at build before any modeling and the coding rule is frozen here: use Mayhew's list through its last fully covered Congress, and do not extrapolate past it. ## 3. Outcome measures **Primary DV:** count of significant enactments per Congress from Mayhew's updated list, using his combined ("Sweep Two") significant-legislation total. **Robustness DVs (independent codings, used where they overlap the span):** - Clinton and Lapinski, *Measures of Significant Legislation*, item-response significance, 1877 to 1994. Overlaps only the pre-1994 portion. This tests whether the divided coefficient survives a completely different significance methodology in the overlap. - Stathis / Congress Project *Important Legislation*, to 2014. - Ansolabehere, Palmer, and Schneer independent full-history counts, if the per-Congress series is recoverable. A claim is robust only if the divided coefficient agrees in sign and rough magnitude across the measures where they overlap. Where they do not overlap (the polarization era), the evidence rests on Mayhew alone and we say so plainly. ## 4. Key predictor Divided government, coded 1 when the president's party does not control both chambers, 0 when it controls both (Mayhew's definition). We adopt Mayhew's own published divided/unified classification for the Congresses he covers so the predictor matches his series, and cross-check it against the executive-orders study's control table (maker-checker). Any discrepancy is reported, not silently reconciled. Mid-Congress switches (for example the 2001 to 2002 Jeffords realignment, 107th Congress) follow Mayhew's coding for consistency; for any Congress we must code ourselves, party control is fixed as of the January organizing of the Congress, documented in an appendix. ## 5. Controls **Primary (parsimonious) model:** divided + linear time trend + war. - **Time trend.** Congress number, centered. Mandatory, not optional. Landmark-law counts have declined secularly and divided government became more common after 1994, so naive pooling would attribute to "divided" what is really "recent." Without this term the estimate is worthless. - **War.** Indicator = 1 if the United States was in a Correlates-of-War-coded interstate or intrastate war during either year of the Congress. The exact per-Congress coding is listed in an appendix so it is auditable. **Full model (robustness):** adds - **Post-election Congress** (honeymoon), = 1 for the Congress convening the January after a presidential election. - **President's-party seat share**, House and Senate. - **Democratic president**, = 1. The divided coefficient must agree across the parsimonious and full models for any robust claim, mirroring the pooled-versus-fixed-effects honesty of the executive-orders study. If it moves, we report the fragility rather than picking the friendlier specification. ## 6. Estimation - **Primary:** negative binomial regression, incidence-rate ratio and 95% CI reported for divided. Poisson as robustness, with the overdispersion test reported. - **Serial correlation:** the series is temporal, so residual autocorrelation is likely. Pre-register a serial-correlation check and HAC (Newey-West) standard errors, mirroring the pendulum study. - **Stability:** leave-one-Congress-out and leave-one-decade-out re-estimation of the divided coefficient. If dropping a single Congress or decade flips the sign, the effect is not robust and is reported as such (this discipline caught the super PAC and pendulum fragilities). ## 7. Secondary hypothesis (registered as underpowered) **H2:** the gridlock effect emerges or intensifies with polarization. Operationalized as an interaction divided × polarization, with polarization measured as the voteview DW-NOMINATE between-party distance per Congress (continuous, principled, already used in the committee study). Registered limitations, stated before data: - Few high-polarization divided Congresses exist, so this test is underpowered by construction. A null is not evidence of no effect. - A significant interaction is reported as suggestive only unless it survives the same measure-agreement and leave-one-out discipline as the primary. ## 8. Completeness gate - Every Congress in the span must carry a count from the primary measure or the pipeline halts loudly. A Congress with genuinely zero landmark laws is a real zero; a Congress missing from the source is ignorance and halts. No silent zero-filling (the Federal Register lesson). - Endpoint verified before modeling; no extrapolation past Mayhew's last covered Congress. - Divided/unified coding cross-checked against the executive-orders table; discrepancies reported. ## 9. Pre-registered predictions - **H1 (popular claim under test):** divided government reduces significant enactments. Registered expectation: null (divided is approximately equal to unified), consistent with thirty years of prior work. - **H2 (secondary, underpowered):** positive divided × polarization interaction. Registered expectation: null, with a real interaction being the newsworthy result if it appears and survives. ## 10. What would change the conclusion - A divided coefficient significant and negative in the primary model, stable across the parsimonious and full specifications, agreeing across the independent measures, and surviving leave-one-out, would support the gridlock claim. Anything less is reported as a null or as fragile. - If the interaction is significant but the measures cannot corroborate it in the modern era (because they do not extend there), the finding is reported as suggestive and measure-limited, not as established. ## 11. Known limitations, stated up front - The independent significance measures do not overlap where it matters most. Clinton-Lapinski stops at 1994, so the both-ways-agreement design is weakest in exactly the polarization era the secondary hypothesis targets, leaning mostly on Mayhew plus Stathis-to-2014 there. - "Significant" is a coding choice. We freeze the primary measure now and treat measure choice as a robustness axis, not a tuning knob. - n is small (roughly 38 Congresses). The primary is adequately powered for a main effect; the interaction is not, and we say so rather than reading a null as proof.