What the Data Shows: Separating Weather from Outage
This chapter examines how the resilience study separates a hurricane's broad disruption from the narrower payment shock created by an outage. The key evidence is not a single spending line or cash percentage, but the timing and comparison of several outcomes: outages, store activity, expenditures, and payment composition. Reading those patterns carefully shows what the data support about digital-to-cash substitution—and what remains bundled with physical damage, mobility, and store operations.
What timing can tell us
An event study lines up observations by their distance from a common event. Instead of asking only how much a county spent in a given month, it asks how spending and payment behavior changed in the days before and after a hurricane. The approach makes the sequence visible: preparation before the event, disruption around impact, and recovery afterward. It is useful for mechanism because a payment failure should appear at a different point in time from a household's decision to stockpile.
The first pattern is in the infrastructure itself. Hurricane-related outage intensity rises sharply when the storm arrives, reaches its highest level during the first few days, and remains elevated for more than a week. Tropical storms also produce outages in the study, but the reported response is smaller. This timing matters because an outage that lasts for days is not just a brief inconvenience at a checkout terminal. It can alter several rounds of purchases, the availability of stores, and the ability of households to replenish whatever liquidity they used.
The second pattern is spending. Store expenditures rise before a hurricane and fall sharply when the storm hits. The pre-event increase is consistent with stockpiling: households bring forward purchases when they expect difficult travel, store closures, or supply problems. The post-event decline has more than one possible source. Some stores close, so there are fewer places where a purchase can occur. Among stores that remain open, spending per store also falls. Reduced mobility and local damage can keep customers away, while payment disruption can prevent some customers who do arrive from completing an electronic transaction.
The decomposition into active stores and spending per operating store is important. If only the number of operating stores fell, the evidence would mainly describe physical access to retail. If only spending per open store fell, a payment or demand channel would appear more directly. The study finds movement in both margins. That does not isolate the exact contribution of electronic-payment failure, but it prevents a misleading interpretation in which every lost sale is attributed to a single cause.
The third pattern is payment composition. Before the storm, the cash share of store sales falls. After impact, it rises sharply and stays higher for more than two weeks in the reported event-time pattern. Credit and debit shares decline after the event, while credit use rises especially before it. The sequence fits the mechanism established earlier: households may use cards to finance stockpiling, preserve cash for the disruption window, and then turn to cash when electronic methods become less reliable.
Payment composition adds information that total spending cannot provide. A fall in sales could mean people need fewer goods, cannot reach stores, or cannot pay. A rise in the cash share suggests that at least some transactions are moving away from cards or other electronic methods. It does not by itself prove that every shift was forced by a failed terminal. Households may also spend cash that they accumulated in advance, even when their local electronic network continues to function.
The resilience study checks whether the pattern is confined to convenience stores. Broader county-level card aggregates rise before hurricanes and fall sharply on impact, and challenger-bank debit data show a similar post-event decline. These additional sources do not solve every identification problem, but they make it less likely that the timing is an artifact of one retail segment. Convenience stores are especially relevant for emergency purchases such as fuel, water, batteries, and basic supplies; broader card data show that the broad spending response is not unique to that setting.
The order of these observations matters. Outage intensity is the infrastructure measure: it tells us whether the payment environment was disrupted and how long the disruption persisted. Store activity and spending show the economic outcome, but combine access, demand, mobility, inventory, and payment constraints. Payment composition then helps identify a channel within that outcome. If expenditure and cash share move together, the interpretation is stronger than if either measure is viewed alone, but the measures still do not become interchangeable. A cash-share increase is not a direct count of failed transactions, and a spending decline is not a direct count of payment failures.
An ordinary before-and-after comparison would lose this separation. Imagine comparing a county's spending during a hurricane week with its spending in a calm week. A large decline could reflect the storm's physical damage, the fact that people purchased supplies earlier, or a loss of electronic payment access. It would be impossible to know how much of the change would have occurred if the county had experienced the same weather but retained power and communications. The neighboring-county design does not create that unobserved counterfactual perfectly, but it moves closer by comparing places in the same regional weather environment.
A useful thought experiment makes the logic concrete. Suppose two nearby counties face the same storm warning. Households in both counties buy more supplies beforehand. In one county, electricity and payment connections remain available; in the other, an outage interrupts electronic transactions. If both counties then show the same large spending collapse and the same cash shift, the outage would add little explanatory power. If the no-outage county returns near its usual spending path while the outage county experiences a sharp decline and stronger movement toward cash, the difference is consistent with the outage changing what transactions can be completed. It remains a comparison, not a laboratory experiment, because the counties may differ in damage and infrastructure. But it is a more informative test of the payment channel than treating every storm-exposed county as equivalent.
The event-time pattern is therefore informative in a specific way. It supports a sequence in which households prepare before the storm, spending contracts when the event arrives, and payment composition moves toward cash afterward. It does not tell us that payment failure caused the entire contraction, nor does it show that every household went through every stage. The interpretation becomes sharper when counties facing a similar storm environment are compared by whether they actually experienced an outage.
The comparison that changes the interpretation
A before-and-after comparison around a hurricane cannot by itself distinguish the storm from the outage it causes. The neighboring-county design addresses that problem by comparing places exposed to the same broader weather environment but differing in whether an outage occurred. The comparison is not a perfect experiment. It is valuable because it holds more of the regional shock in common than a comparison with an unaffected county or an ordinary pre-storm day.
The no-outage counties show the response that can occur when households face the storm environment but the payment infrastructure remains available. Spending rises before the event, which is consistent with preparation. Afterward, there is little evidence of the large or persistent spending decline seen in outage counties. The cash share nevertheless rises. That combination is consistent with a precautionary story: households build or preserve cash before the storm and later spend some of it, even though digital payments remain usable.
The outage counties show a different pattern. Their expenditures fall sharply at the event, and the movement away from digital methods toward cash is stronger. Because these counties are neighboring places exposed to the same general storm conditions, the contrast makes a payment-system channel more plausible. The weather event raises preparation needs in both groups; the outage additionally narrows the set of payment methods that can be used. The result is not merely “bad weather causes people to use cash.” It is that infrastructure failure changes the feasible choice set.
This comparison also clarifies why a rising cash share is not a single mechanism. In a no-outage county, an increase can reflect the spending of precautionary balances. In an outage county, it can reflect substitution after cards or other electronic methods become unavailable. Both processes may operate in the same place, but the expenditure response helps distinguish them. Drawdown alone need not reduce total spending sharply if stores and electronic payments remain available. When an outage is added, cash can preserve some transactions while the overall amount spent still falls because other transactions are blocked.
The result has a useful asymmetry. Cash substitution can be evidence that cash is helping, but a higher cash share does not mean the economy is fully resilient. In an outage county, total spending can decline while the cash share rises. The denominator has changed because the electronic part of the payment system has contracted, and physical disruption may have reduced the number of available purchases as well. Cash can protect a subset of transactions without restoring the whole level of commerce.
The neighboring-county comparison is stronger than a simple temporal contrast, but its limits are material. Outage realization is not necessarily random. Counties may differ in grid design, retail availability, household cash holdings, storm intensity, damage, or the speed of restoration. Neighboring places can share a weather system without sharing the same infrastructure or exposure. A county with an outage may also suffer different physical damage, and that damage can affect both stores and payment access.
Those limitations do not erase the result; they determine its scope. The evidence supports the conclusion that outages are an important channel through which extreme weather affects spending and payment composition. It does not prove that every dollar of the expenditure decline comes from failed electronic payments, or that cash alone would have prevented the decline. A careful statement is therefore conditional: when stores remain open and cash is available and accepted, cash can sustain transactions that a disrupted electronic route would otherwise block. When stores close or goods cannot be obtained, the payment method is no longer the binding constraint.
The study's multiple datasets reinforce this conditional reading. Store transactions reveal payment instrument and operating-store changes. Card aggregates show whether a similar timing appears beyond one retail format. Household purchase data connect spending and payment choice to household characteristics. Surveys speak directly to failed preferred methods and cash fallback. Each source covers a different part of the mechanism, and none should be asked to answer a question it cannot observe.
This is also why the phrase “cash became more important” needs a denominator and a comparator. More important than what? More common than before the storm, or more important relative to a digital method that stopped working? In the no-outage comparison, cash's share can rise because a buffer is spent. In the outage comparison, it can rise because electronic acceptance fails. The same direction of change carries different information depending on the local failure condition.
A sound interpretation therefore has three layers. The event study establishes timing: preparation precedes the storm, spending falls at impact, and cash share rises afterward. The decomposition identifies coexisting channels: fewer active stores, lower spending at operating stores, and a change in payment composition. The neighboring-county comparison assigns greater weight to the outage mechanism because the sharpest changes occur where the digital infrastructure fails, not merely where the storm is observed.
What this evidence still does not answer is whether cash availability changes the outcome for a household that would otherwise be unable to complete a purchase. Aggregate payment shares can show substitution, but they cannot fully distinguish a cash-preferring household from a card-preferring household forced to use cash. To answer that question, the analysis has to move closer to household liquidity: who has cash available, who completes a transaction when digital methods fail, and whether access to cash is associated with a smaller fall in consumption.