Skip to main content

Welcome to a Grasp lesson

Create a free account to keep studying and create your own study path.

Lesson 13

Correlation vs. Causation in Political Connections and Business Rewards

Recommended resources

Welcome back. The previous lesson showed how to locate records of a political connection, a public decision, and a business reward. Finding all three is an important start, but it leaves a harder question: did the connection help produce the reward? This lesson develops a way to separate an observed relationship from causal evidence, while keeping legitimate explanations for an award in view. Allow about 40 minutes for the lesson, video, and selected reading.


A connection and an award can coincide for several reasons

Suppose a report finds that politically connected suppliers win more public contracts than other suppliers. That is an association: connection status and contract success occur together in the observed data. A causal claim goes further. It says that, for at least some suppliers, the connection changed what they would otherwise have won.

The distinction matters because firms do not acquire connections at random. A large supplier may have both the capacity to win complex contracts and the resources to hire lobbyists. Past success may make its representatives more likely to meet officials. An agency might repeatedly buy from a supplier because its equipment works with existing systems. None of those possibilities proves the process was fair; each is an alternative that the favoritism claim must confront.

A short video from the Winton Centre shows how easily an association can become a stronger claim as it passes from research into a headline. Its examples come from other fields, but the same change in wording occurs when “connected suppliers won more” becomes “connections made them win.”

Correlation vs Causation: A Brief Guide To Communicating Research

Watch “Correlation vs Causation: A Brief Guide To Communicating Research” from the Winton Centre for a compact account of the distinction and how to report it accurately.

Watch a common cause behind an apparent relationship. Continue through headline drift, noticing the change from a reported association to a causal claim. In study limits, focus on why simply observing two groups is different from assigning an intervention. Finish with careful wording; the distinction will help you state a procurement finding without claiming more than the evidence supports.

For the supplier example, several explanations can produce the same pattern:

  • Favoritism: an official uses a connection to shape specifications or selection in a supplier’s favor.
  • A common cause: established suppliers both win more tenders and have more contact with government because they are larger or more experienced.
  • Reverse direction: a supplier’s earlier contracts lead it to build political relationships, rather than those relationships causing the earlier contracts.
  • Both directions: an established supplier gains access through past success, and that access later helps it obtain further advantages.
  • Chance or selective reporting: a striking pattern may look less distinctive when all eligible firms, agencies, and time periods are included.

These are possibilities to test, not interchangeable excuses. Timing can rule out some claims—for example, a connection first formed after an award cannot have caused that award through the alleged channel. Timing alone cannot establish the remaining explanation.



The missing comparison: what would have happened without the connection?

The causal difficulty is that we cannot observe the same supplier, tender, and moment both with and without a political connection. The unobserved result is the counterfactual. Researchers seek a comparison that gives a credible estimate of it.

The outcome must also be specified. “Received a large contract” is not the same as “earned an unusually high profit.” An award amount is a recorded commitment for goods or services, not a measure of excess return. For a claim about award decisions, a useful outcome might be the share of eligible bids that win, alongside bid prices, technical scores, and the rules under which bids were judged.

The following table from J-PAL compares evaluation methods by the assumptions needed to interpret them. Read only the four introductory methods specified below; the later statistical methods are not needed for this lesson.

Impact Evaluation Methods Table

Read J-PAL’s “Impact Evaluation Methods Table” to see why a comparison can look persuasive yet depend on a demanding assumption. Translate its “program participants” into connected suppliers and its “outcomes” into contract success.

In the PDF’s methods table, begin with the “Randomization / Randomized Evaluation/ Randomized Control Trial” row. Read the comparison, especially why assignment makes groups more comparable. Under “Basic Non-Experimental Comparison Methods,” read the “Pre-Post” row from its description through its warning about outside changes. Then read the “Simple Difference” row, from the group comparison through its discussion of selection into a group. Finally, in “Differences in Differences,” read from the method through its assumption about how both groups would have changed without the intervention. The PDF is a table: follow each named row across its description and assumptions columns rather than reading straight down one column.

Random assignment would make a comparison powerful because, before the intervention, it tends to distribute both observed and unobserved advantages across groups. Political ties are rarely assigned this way, so an investigation normally relies on observational evidence.

Two tempting comparisons are weaker than they first appear. A before-and-after comparison of one supplier may mistake a market-wide rise in purchasing for an effect of its new connection. A connected-versus-unconnected comparison after an award may mistake differences in experience, eligibility, or bid quality for an effect of connection. Even comparing suppliers with similar recorded characteristics leaves open consequential differences that the records do not measure.



A worked comparison—and what it cannot settle

Consider a hypothetical study of suppliers bidding on similar equipment tenders. One group gained a documented political tie between two periods; another did not. The reported outcome is the proportion of eligible bids that won:

Supplier group Before After Change
Gained a documented tie 20 of 100 bids (20%) 30 of 100 bids (30%) 10 percentage points
Did not gain a tie 18 of 100 bids (18%) 23 of 100 bids (23%) 5 percentage points

A simple report might emphasize that the newly connected group’s success rate rose by 10 percentage points. But the comparison group’s rate rose too, perhaps because agencies bought more equipment or tender conditions changed generally. A difference-in-differences comparison subtracts that shared increase:

The resulting five-percentage-point gap is more informative than either the after-period gap or the connected group’s increase alone. It is not automatically “the effect of connections.” To interpret it that way, the groups would need to have followed sufficiently similar trends without the new ties. That assumption becomes less credible if, at the same time, the newly connected firms introduced a better product, entered less competitive tenders, or became eligible for contracts the other firms could not bid on.

Earlier periods can help: markedly different success-rate trends before the ties formed would weaken the comparison. Similar earlier trends would support it but would not prove that no new difference emerged later. It also matters why firms gained ties. If firms expecting a surge in relevant contracts were especially likely to seek political contacts, the timing reflects their expectations as well as any subsequent influence. With only 100 bids per group in each period, ordinary variation is another consideration; the table alone does not establish how precise the estimated gap is.

A careful description is therefore: In this hypothetical comparison, the success rate increased five percentage points more for firms that gained a documented tie than for the comparison firms. Whether the tie caused that difference depends on the comparability of the groups and other changes during the period.



Use the decision record to probe the causal explanation

Numerical comparisons address whether an unusual reward pattern remains after some alternatives are considered. The records from the previous lesson can address a different part of the claim: how a connection might have affected a particular decision.

Imagine that a connected supplier wins after an agency narrows a technical specification. An influence explanation gains support if records establish that the relationship existed before the change, the changed requirement excluded otherwise capable competitors, and the documented rationale does not adequately explain the restriction. A public-interest explanation gains support if contemporaneous technical records show why compatibility or safety required the feature, comparable suppliers could meet it, and the announced evaluation criteria were applied consistently. The requirement’s effect on competition and the adequacy of its justification matter more than the supplier’s political identity alone.

Evidence must be matched to the proposition it supports. A lobbying filing can substantiate reported activity in the relevant period, but not an official’s reason for adopting a requirement. A solicitation amendment can establish that the rule changed, but not why. An award notice can identify the winner, but not whether the contract was unusually profitable. Consistent timing, a plausible decision pathway, comparisons with qualified rivals, and contemporaneous reasons are stronger together than any one record is alone.

This approach avoids two opposite errors. An observed association need not be dismissed merely because there was no experiment: carefully chosen comparisons and decision records can support a cautious causal inference. Equally, an association that survives one comparison is not proof of favoritism. Unmeasured bid quality, agency needs, and the selection of firms into political relationships may still matter.



Conclusion

Correlation establishes that political connections and business rewards occur together in the observed cases. Causal evidence must make the counterfactual more credible: it checks timing, compares genuinely comparable firms or decisions, considers changes affecting both groups, and examines the public decision through which an advantage allegedly arose. The warranted conclusion may be strong, tentative, or unresolved; its wording should reflect that difference.

The next lesson will place the favoritism and public-interest explanations side by side in a structured evidence table, making both supporting evidence and remaining uncertainty visible.