Predicting Homelessness with Heuristics
The example compares heuristics and regression for ranking homelessness rates across 50 U.S. cities using six cues. It distinguishes performance on the training data from prediction on held-out data; the supplied passage describes the comparison without reporting its results.
THE IDEA
Predicting Homelessness
The comparison uses data from William Tucker’s article (1987) exploring homelessness in 50 U.S. cities. The six candidate cues are unemployment, poverty, housing vacancy, rent control, public housing and average temperature.
The proposed relationships involve both the ability to pay for housing and its availability. Vacancies may affect rental options; rent-control policies and public housing may affect supply and affordability. Temperature might affect how tolerable sleeping outside is and where people experiencing homelessness live. These are proposed explanations for associations, not causal findings established by the prediction exercise.
We ask our heuristics to use these six (dichotomized) cues to predict homelessness rates in the 50 cities. The heuristics are required to choose the city with more homelessness for all 50 × 49*/*2 = 1225 pairs of cities. Regression uses the matrix of cue values to derive optimal weightings of the cues (possible causes). There are two types of competitions. In the first competition, the test set is the same as the training set (from which a strategy learns the information it needs, such as the weights of the cues). In the second, more realistic competition, the test set is different from the training set (also known as cross-validation). The second competition can reveal how much a heuristic overfits the data. Only the first type of competition was studied in Gigerenzer and Goldstein (1996).
When to use it
Predicting homelessness rates in cities with limited data using a few key economic indicators.
Testing whether a proposed decision rule predicts need accurately before using it in resource-allocation decisions.
Distinguishing variables that help prediction from factors shown to cause changes in a social outcome.
Comparing the effectiveness of simple decision rules versus complex models in policy-making.
FROM THEORY TO EVERYDAY LIFE
In practice
Applying Heuristics to Homelessness Prediction
Tucker’s article (1987) supplied data on six possible correlates of homelessness in 50 U.S. cities. The later comparison used these data to test simple decision rules against regression when choosing which of two cities had a higher homelessness rate.
Heuristic vs. Regression in Real-World Decision Making
A policy team could compare a simple rule with a regression model on data neither method was trained on. The passage explains why this is a more realistic test than evaluating either method only on its training data; it does not establish that a particular rule is accurate enough for allocating assistance.
A MOMENT TO REFLECT
Put it to work.
How might simple heuristics outperform complex regression in predicting real-world outcomes like homelessness?
What are the advantages of using heuristics in decision-making under uncertainty, especially when data is limited?
Can you identify a situation in your own field where a heuristic could be more practical than a statistical model?
Often confused with
References
- Article on homelessness causes — William Tucker (1987)
- Heuristic and regression comparison study — Gigerenzer and Goldstein (1996)
CONTINUE THE CONVERSATION
Source texts
- Heuristics and Biases The Psychology of Intuitive Judgment (Thomas Gilovich, Dale Griffin, Daniel Kahneman)