Reference

Analysis Tools

A method is not a neutral instrument. It is a set of assumptions about how the world works, chosen because it suits a particular question and a particular kind of data. Change the assumptions and the answer can move.

So these are short notes on the tools behind the research: what each one does, in plain words, why it suited the question, and an honest line about what it cannot do. If you want to argue with a finding, start here, because the argument is almost always about an assumption rather than about the sums.

None of this is a new interest. In 2012 I published a working paper with Trade and Industrial Policy Strategies appraising how impact assessment methodologies were being chosen for South Africa's Community Work Programme, which asked the same question this section does: not whether a method is sound, but whether it is the right one for what is being asked. It is not linked here because TIPS appears to have dropped it in a site rebuild and the address on my own CV no longer resolves.

Maize trade in Southern Africa

Master of Science in Agricultural Economics, University of Pretoria, 2012. What actually determines how much maize moves between SADC countries.

  1. The Gravity Model

    Borrowed from physics, and the name is the clue. Big economies pull harder, and distance pushes trade apart.

  2. Panel Data

    One year of data cannot tell a lasting pattern from a fluke of that year. Following the same units over time can.

  3. Fixed and Random Effects

    Two ways of handling the things about a country pair that never change. The choice is a judgement, and it is worth stating out loud.

  4. The Tobit Model

    Most countries trade no maize with each other in most years. A zero is a decision, not a small number.

  5. Maximum Likelihood

    Of all the values these numbers could take, which set makes what I actually observed most likely to have happened?

South African exports under AGOA

Master of Commerce in Management Practice, Trade Law and Policy, University of Cape Town, 2015. What South Africa actually got out of AGOA.

  1. Why I Built No Model

    Three modelling approaches were available. I reviewed all of them and used none, and I would rather say so plainly.

  2. Descriptive Trade Analysis

    The least glamorous method here, and the one that did most of the work.

  3. The Trade Intensity Index

    Do these two countries trade more with each other than their size alone would predict?

  4. The Preference Margin

    The number at the centre of every argument about trade preferences, and the one most often left unstated.

  5. Revealed Comparative Advantage

    Let the trade data tell you what a country is good at, instead of deciding in advance.

  6. Indicative Trade Potential

    Match what a country is demonstrably good at against what a market demonstrably wants, then look at the gaps.

Climate-smart agriculture in Zambia

Ongoing research at North-West University. The household work draws on the Water and Soil Accelerator survey of 2,116 farming households, funded by USAID; the supply-side work profiles twenty-six enterprises. Which practices farmers prefer, who adopts, what adoption does to income and yield, and who is selling the technology.

  1. Profiling the Enterprises

    Twenty-six firms selling climate-smart technologies. With twenty-six of anything you learn more by asking than by modelling.

  2. Cochran's Q

    Are these six practices really used at different rates, or is that just how the numbers fell?

  3. McNemar's Test and the Bonferroni Correction

    Only the farmers who did one thing and not the other tell you anything about a preference between them.

  4. The Five Capitals

    Why these variables and not others. A framework that answers the question rather than leaving it to taste.

  5. The Bayesian Multinomial Logit

    Three frightening words, taken one at a time. The third one was forced on me by a category with a single household in it.

  6. Average Marginal Effects

    A model that speaks in log-odds is a model almost nobody can read. This is the translation step.

  7. Endogenous Switching Regression

    Nobody randomised who adopts. So you have to build the world in which they did not.

  8. Double Machine Learning

    Predict the outcome, predict the choice, strip both out, and study what is left over.

  9. Bayesian Double Machine Learning

    The same logic, returning a distribution instead of a single number.

  10. The Causal Forest

    An average answers a funding question. It tells an extension officer with one motorbike nothing at all.

  11. The Assumption About the Neighbours

    Every estimator here assumes your neighbour's decisions do not reach you. In a farming village that is plainly untrue.

  12. The Robustness Checks

    The last job is not to get an answer. It is to establish how much weight the answer will bear.