Syndicate content

The history of randomized control trials: scurvy, poets and beer

Markus Goldstein's picture
The other night I was reading Julian Jamison’s well written and intriguing recent paper: “The Entry of Randomized Assignment into the Social Sciences,” which plumbs history to understand where it is that RCTs come from.   
 
The ideas start with the title.   Julian draws a distinction between random assignment and random allocation.   Random assignment is when the condition (“what was done, when, to whom or what, etc”) is randomly assigned across observations.   Random allocation – which is what I think of when I think RCT – is related but (I think) a subset of random assignment.   Julian argues that “randomized assignment serves a deeper purpose than simply impartially dividing a sample into subsamples, and it may apply even when causality is not a central concern: it guarantees us that any two observations we collect are comparable (in expected terms) across dimensions other than those we know about and vary in a controlled manner.” 
 
So this notion of random assignment is what gives birth to RCTs.   Where did the idea of a control group come from?   Julian charts the first written idea back to 1364 and a letter from the poet Petrarch in a letter to Boccaccio (also a poet and writer) where Petrarch argues for comparing a group of “a hundred or a thousand men” who follow medical advice with the same size group that doesn’t.   And he bets one group will survive.  
 
Putting the idea into practice comes later when a Scottish naval surgeon, James Lind, whom on a voyage in 1747 is trying to cure scurvy.  He compares two treatments across six pairs of sailors. One treatment is two oranges and one lemon a day (which, in case you are a landlubber, works).   As Julian points out, in setting up the experiment Lind was trying to make sure that the men in the different treatments were as similar as possible:  “The cases were as similar as I could have them.   They all in general had putrid gums, the spots and lassitude, with weakness of their knees.   They lay together in one place, being a proper apartment of the sick in the fore-hold; and had one diet common to all.”
 
The first actual random assignment to treatment and control comes out of Nuremburg in 1835 and an effort to measure the impact of homeopathic medicine (yes, it’s been going on for awhile).  Here a “society of truth-loving men” randomly assign 100 volunteers randomly to receive water versus water with a salt-based treatment.   This trial was actually double blind – so another idea that started quite early.   (In case you were wondering – no differences across treatment and control).
 
In 1884, we get the first randomization in the social sciences.   The (among other things) psychology researcher Charles Pierce was trying to examine a hypothesis that people’s perceptions wouldn’t let them distinguish really small differences in weight.  Working with a student of his (another pattern that seems to have started early), they randomized whether the subject (one of the two of them) got the base weight or the supplemented weight first.   As Julian points out, Pierce was randomizing over stimuli rather than subject with the goal of comparing “like with like.”  So random assignment, but not random allocation. 
 
In the 1920s things really start to heat up.   One interesting thing is that research in medicine, psychology, and agriculture start implementing proper randomized control trials at the same time.   Over at the Rothamsted agricultural research station in the UK, Ronald Fisher is sorting through data and running randomized field experiments (publishing the first with Eden in 1927).   Meanwhile in medicine, Colebrook (1929) is drawing lots to administer radiation to children (as Julian puts it: “not as bad as it sounds”).    And in psychology, folks working in education were randomly assigning folks to groups (e.g. Shaffer in 1927).   
 
Economics enters the fray shortly thereafter.  The earliest work seems to be laboratory experiments on indifference curves by Thurstone (1931), which was published in a psychology journal.   Chamberlin’s   1948 laboratory experiment on demand and supply curves is often cited as the first laboratory experiment in economics.  
 
But it is the political scientists who first take this to the field.   For the US presidential election of 1924, the political scientists Harold Gosnell sent residents of some districts in Chicago a postcard to increase registration and compared those to other districts where none were sent (Julian writes that “results were encouraging”).   Julian notes that there is some debate over whether this was randomized or not, but that Gosnell “clearly understood the need for a rigorous control group in order to isolate causal factors, which is what kept driving scholars toward randomization.   The timing is not coincidental here: social policy, educational psychology, clinical medicine, and agriculture all used randomized assignment with a few years of each other.”  
 
Following this is a growth in trials starting from a study on counseling and treatment for juvenile delinquents in the 1940s and a very large study of the effects of the polio vaccine in the 1950s.   What struck me about some of the studies during this growth period was how some of the issues in fashion in economics now show up back then.    For example, Julian discusses a 1964 study designed for measuring spillovers (in family planning), a 1972 study looking at financial incentives for heath workers, and a 1971 study looking at extrinsic versus intrinsic motivation.   
 
The first field experiment in economics shows up in 1960s when Heather Ross (a grad student!) looks at the impacts of negative income tax.   Development economics wasn’t that far behind – in 1974 there was an experiment on teaching math over the radio in Nicaragua (it worked). 
 
 
These are only some of the highlights – for a good time (and more graphic detail of disease and crazy cures as well understanding what the t-statistic, matching and Guinness have in common) I recommend reading the whole 27-page paper.   And who knows, we might get a sequel; as Julian writes in his conclusion “the recent (après 2000) rise of RCTs in economics could almost be said to embody a fourth phase, although one that no longer concerns the entry of the concept into the intellectual environment.  How and why that happened, as well as the strenuous pushback that it has received, is a tantalizing arena for future study.” 
 

Comments

Submitted by Marlaine Lockheed on

The Nicaragua radio education experiment was only partially done by an economist (Julian’s dad); the others were Barbara Searle, a biologist, and Patrick Suppes, a philosopher.

Submitted by Julian Jamison on

Hi Marlaine :) Yes Pat Suppes appears prominently in the article (identified as a philosopher!), starting with laboratory work in the 1950s...

Add new comment