The pay-in-4 file: five studies, one pattern

2026-10-21 · spending, credit, meta-analysis · about the collection: BNPL / Pay-in-4: the spending button

The first post in this series leaned on one working paper. That felt thin, because it was. So the collection grew: five studies now sit in the BNPL file, and they don't share a country, a dataset, or a method. Bank transactions in the US, retailer transactions in the US, a randomized experiment inside a billion-user wallet in China, credit card records in the UK, and a Federal Reserve consumer survey. When five different microscopes on the same checkout button mostly agree, the finding stops being "a result" and starts being "the way this works."

(The full collection lives here — every study with its digest and source PDF, plus the papers that belong in the file but can't be legally fetched yet.)

TL;DR: Across three independent causal designs, turning on pay-in-4 raises spending — about +$130 on impact in bank data, +9 percentage points purchase frequency and +10% basket size at one retailer, +5.6% of wallet spending in Chinese transaction data. The costs are just as consistent: overdrafts up 20% relative, missed-installment fees of $7–10, and roughly one in five UK credit cards now carrying BNPL balances at ~20% APR. The one big disagreement — Chinese users barely default — points the finger at stacking, not installments.

Same button, five microscopes

That's the honest spread of designs. No pooled effect size here — a DiD on bank data and a field experiment in a super-app don't average into anything meaningful. What they can do is agree or disagree, and the pattern is the story.

Where five studies agree

The button raises spending beyond the purchase it finances. The anchor study's spillover — spending rising well past the installment purchase itself — is the lonely result no more: the first post told that story, so here's just the cross-check. Maesen and Ang see the same pattern inside a single retailer: adopters buy more often, and their baskets grow, well past the installments they actually take. Bian's experiment shows the same direction with the cleanest possible identification: hand users BNPL access at random, spending rises within two months. Three designs, two continents, one direction.

The button works hardest on the people with the least slack. The spending lift concentrates among younger, lower-income, and credit-constrained users in every dataset that can see demographics — and Akana's survey explains the funnel: adoption deepens fast among people already using it, while only 2.5% of nonusers say they'll ever start.

The "0% interest" label hides a fee schedule. The anchor study found overdraft odds rising ~20% relative; Akana's survey independently documents missed-installment penalties of $7–10 at Afterpay, Klarna and Zip, capped near 25% of the purchase. Two different data types, same conclusion: the real interest rate arrives wearing a "no fees" t-shirt.

Where they disagree — and why it matters

The sharpest split in the file is Bian vs everyone else. In the Chinese wallet data, default sits below 1%, and about two-thirds of users who rely on BNPL as their only credit pay in full every cycle, moderating their borrowing the moment interest appears. Western data shows overdrafts, revolving, and debt stacking instead.

The collection's read: the difference isn't the installments, it's what's underneath them. In the Chinese sample, BNPL is often the only formal credit a user has — there's nothing to stack it on, so it behaves like digital cash. In the UK data, 19.5% of active credit cards already carry BNPL transactions, and that's where the harm lives: 0%-interest, few-weeks debt refinanced onto ~20% APR balances that minimum payments stretch for decades. Guttman-Kenney's regulatory question — should you be allowed to refinance unsecured debt with more unsecured debt? — is the right one, and it's really a question you can ask at your own checkout.

There's a smaller disagreement worth naming. Akana's users mostly describe BNPL as convenience and budget management — 43.8% say it helps them manage finances better. Every transaction dataset in the file shows those users spending more after adoption. Both can be true and still be the trap: the tool that makes spending feel manageable is doing the making.

What nobody has tested yet

• Debt stacking, directly. No observer can see one consumer's Klarna + Afterpay + Zip balances side by side. The CFPB's market report documents the practice; no causal paper has measured what stacking does to default. This is the file's biggest hole.
• Long-run wealth effects. Every causal estimate here is weeks-to-months. Whether the +$60/week persists for years — and what it does to net worth — is unmeasured.
• Killing the effect at checkout. Maesen's experiments hint the lift shrinks when the full total is made salient, but that's tested in purchase scenarios, not a real checkout. A retailer that A/B-tested "here's your $600 total" against "4 × $150" would own the most useful result in the literature.
• Canadian data. Everything transaction-based here is US, UK, or Chinese. Direction probably travels; exact dollars may not.

One honest caveat before the advice: this collection's peer-reviewed backbone is Maesen & Ang (Journal of Marketing) and Guttman-Kenney et al. (JBEF) — so the claims "installments raise spending at the retailer level" and "BNPL gets refinanced onto credit cards" stand on peer-reviewed support. The headline dollar figures in the anchor study (Di Maggio) and the Chinese experiment (Bian) come from working papers, which is why they're labeled supplements and why the collection page tags every member. The survey is self-reported. The convergence across independent designs is the reason the pattern is credible anyway.

What to actually do

The anchor post's checklist (treat the button as a spending decision, log the schedule, automate the alternative) still stands — that's the per-transaction playbook. This file adds the rules you can only see from five studies at once:

• The stacking rule. The harm case isn't installments, it's installments plus revolving debt. Pay-in-4 charged to a card that carries a balance is the single mechanism every dataset here points at. Never route it through one.
• Context decides, not the button. Bian's users treat BNPL like cash because it's their only credit. If installments are your third liquidity source rather than your only one, you're in the Western pattern, not the Chinese one. Count your layers honestly.
• Use the salience counter while it's free. Studies 2–4 in Maesen & Ang hint the lift deflates when the full total is visible. Reading "$600" out loud instead of "4 × $150" costs nothing and matches the one lever the literature has tested.
• Watch what would change this file. A causal debt-stacking study, a peer-reviewed replication of the headline magnitudes, or a retailer A/B-testing total-price salience would each move members between the strength labels above. Until then, the labels are the honest ceiling.

How strong is the evidence?

Scale: Strong = replicated across multiple peer-reviewed studies · Moderate = one solid peer-reviewed study or consistent findings in limited settings · Suggestive = working paper, small samples, or not yet replicated · Thin = non-peer-reviewed source or conflicting evidence.

Next steps

• The BNPL / Pay-in-4 collection — the working view: five member papers, plus the next-up list of studies we can't legally fetch yet (the Management Science pain-of-paying experiment is the one to watch).
• The original post: Pay in 4 is a spending button — the anchor study, expanded.
• Every claim above links its skim view — digest beside the full open-access PDF, one tap from the source.