← PLSC 340 · The Politics of International Development Kenya Lab

Politics of international development · Field lab · Week 7

Whose Turn to Eat

The World Bank and the African Development Bank pay for the projects, but Kenya's government helps decide where they go. Change who holds power and how much say the donor keeps, then see which of 210 constituencies get the aid.

Built on Jablonski, R. S. “How Aid Targets Votes: The Impact of Electoral Incentives on Foreign Aid Distribution.” World Politics 66(2), 293–330 (2014).

Who holds power
—
Map shows
The incumbent's rule—
Donor vs government
Donors usually delegate siting to recipient ministries (the Paris Declaration commits them to it). 100% is Kenya as Jablonski describes it. At 0%, the donor picks every site itself.
The government knows which places are poor, and the donor knows less. This sets how closely the donor's guess tracks real need. It only matters for the sites the donor picks.
Jump to a seat
Or tap any hexagon. Every constituency gets one hexagon, so the north looks smaller than it is.

01Test it the way Jablonski did

Every dot is a constituency in the regime you picked. If the incumbent rewards its core, aid rises with its winning margin, and the curve bends upward at the strongholds. If the incumbent courts swing voters, aid peaks near a tied race. Switch the rule and watch the fitted curve.

Aid per person vs incumbent winning margin$ per person per year
What a regression would say

02Is it politics or is it need?

Kalenjin areas got more aid under Moi. But they are also wealthier and better educated than average, so a simple comparison mixes politics with everything else about the place. Jablonski compares each group's aid when its leader is in power with its aid when he is not. Things about the place that stay the same cancel out.

Aid gap, year by yeargroup minus everyone else, $ per person
Difference-in-differences

03Does the aid buy votes?

Jablonski's last test (Hypothesis 5) controls for each constituency's previous margin. Going from the least-aided seat to the most-aided one is associated with about 16 more points of incumbent margin at the next election. Here every seat shifts by that amount, scaled by how its aid compares with the average seat under the allocation you just set.

Closest seats at the next electionincumbent margin, points
margin without the aid effectwith itseat changes hands
Aid effect on margins
16 is Jablonski's estimate (Table 4, Figure 8). He warns it may be too high, because governments may also shift spending during their term in ways the data cannot see.

04What the paper finds

Headline estimates · World Bank & AfDB projects, 1992–2010
The data
210
constituencies, each assigned the aid of the 153 geocoded World Bank and African Development Bank projects located in it. Over $7 billion was committed in all.
$1.38
average aid per Kenyan per year from these two donors.
28%
of Kenyan government spending was paid for by aid in 2009.
Fixed effects (Table 2, Figure 5)
>$1
more per person each year for constituencies that strongly back the incumbent party, compared with strong opposition seats.
≈$0.80
more per person each year for coethnic constituencies, ones whose majority shares the president's ethnicity.
+ curve
The squared margin term is positive. Aid rises fastest among the most loyal seats, so the swing-voter story (H4) is rejected.
Difference-in-differences (Figures 6–7)
+$0.36
more per person each year for Kalenjin seats under Moi. After he left in 2002 they got $0.29 less than the average seat.
Kibaki
Kikuyu seats and strong NARC seats got more aid in 2003–07, and less when Kibaki was not in power.
2008–10
A weaker tilt toward ODM, the party that held the aid-delivery ministries.
Elections (Table 4, Figure 8)
+16 pts
of incumbent margin from going from the least-aided to the most-aided seat, holding the previous margin fixed. It may be an overestimate.
The argument in three steps

1. Donors delegate. Only about 3.5% of donor spending in Africa in 2011 was budget support. Most aid comes as projects, and recipient ministries request, plan and carry them out. Governments know more about local conditions than donors do, which is a good reason to let them choose.

2. Incumbents use what donors don't know. A government that wants to win the next election steers projects to the voters most likely to reward it. In Kenya those are core supporters and coethnics, not swing voters. Ethnic bloc voting and patronage networks make promises to them more credible.

3. It works. Constituencies that get more aid give the incumbent bigger margins next time. Elections are meant to make governments accountable. Here they also give governments a reason to politicize aid. Kenya's 2007 presidential race was decided by about 2 points.

05For discussion

Questions
  1. Take the pen away. Set government say to 0% and the donor's knowledge of need to 20%. The political gap closes. What happens to the aid–need correlation? Is this a better portfolio, and who would argue that it isn't?
  2. Core or swing? Switch to the swing-voter rule and look at the scatter. What shape of curve would have supported Hypothesis 4? Why does Jablonski think Kenyan incumbents favour their core instead?
  3. Why not just compare? Choose “Need only” and open the Kalenjin × Moi panel. The naive comparison is negative even though nobody is playing politics. Why? What does the year-by-year gap need to look like for the difference-in-differences to be believable?
  4. Who is the incumbent? In the power-sharing era, switch control between ODM, PNU and a split. What does Jablonski's coding choice assume about where discretion over aid lives: the presidency, the cabinet, or the ministries?
  5. Sixteen points. Halve the aid effect and look at the 2002 and 2007 elections. Does aid ever decide the outcome? Jablonski says the estimate may be too high. Name a variable that could cause the bias.
  6. Bypass, then what? Donors often respond to capture by going around the government, through NGOs or community-driven programmes. Does that end partisan bias? Compare with the Ghana Lab (Baldwin et al. 2023), where a neutrally allocated NGO programme still had partisan effects.
  7. The 2025 case. If a donor cut its Kenya portfolio by a third tomorrow, which constituencies would lose the most under the rule Jablonski finds? Who would voters blame, the donor or the incumbent?
How this model works

The map is schematic. 210 hexagons, one per constituency, grouped into twelve blocs by majority group. Group sizes, need and vote margins are stylized to resemble the Moi, Kibaki and power-sharing eras. They are not Jablonski's data, and real constituencies were often mixed.

Aid. Each seat starts from the national average of $1.38 per person. The government's share of siting adds a political term. Under the core rule, that is 0.35 × margin + 0.25 × margin², plus $0.35 for coethnic seats. Under the swing rule it is 0.9 × (0.45 − |margin|). The total budget stays fixed, so every targeted dollar comes out of other seats. Need adds $0.30 per standard deviation. The government sees true need. The donor sees a noisy version whose accuracy you set.

Calibration. The core rule puts the gap between the strongest opposition seat and the strongest incumbent seat at about $1 per person, close to Jablonski's estimate. The coethnic bonus is smaller than his $0.80 because his coethnic and margin effects come from separate models and overlap.

Panel and elections. Yearly aid adds fixed random noise to each regime's allocation. The next-election margins come from stylized 2002 (KANU) and 2007 (PNU) results. The aid effect scales with each seat's log aid relative to the average seat, spanning your chosen number of points from the least- to the most-aided seat.