Using covariates to increase the precision of randomized experiments

A simple difference-in-means estimator of the average treatment effect (ATE) from a randomized experiment is a good start, but may often leave a lot of additional precision on the table. Even if you haven’t used covariates (pre-treatment variables observed for your units) in the design of the experiment (e.g., this is often difficult to do in streaming random assignment in Internet experiments; see our paper), you can use them to increase the precision of your estimates in the analysis phase. Here are some simple ways to do that. I’m not including a whole range of more sophisticated/complicated approaches. And, of course, if you don’t have any covariates for the units in your experiments — or they aren’t very predictive of your outcome, this all won’t help you much.


Prior to the experiment you could do stratified randomization (i.e. blocking) according to some categorical covariate (making sure that there there are same number of, e.g., each gender, country, paid/free accounts in each treatment). But you can also do something similar after: compute an ATE within each stratum and then combine the strata-level estimates, weighting by the total number of observations in each stratum. For details — and proofs showing this often won’t be much worse than blocking, consult Miratrix, Sekhon & Yu (2013).

Regression adjustment with a single covariate

Often what you most want to adjust for is a single numeric covariate,1 such as a lagged version of your outcome (i.e., your outcome from some convenient period before treatment). You can simply use ordinary least squares regression to adjust for this covariate by regressing your outcome on both a treatment indicator and the covariate. Even better (particularly if treatment and control are different sized by design), you should regress your outcome on: a treatment indicator, the covariate centered such that it has mean zero, and the product of the two.2 Asymptotically (and usually in practice with a reasonably sized experiment), this will increase precision and it is pretty easy to do. For more on this, see Lin (2012).

Higher-dimensional adjustment

If you have a lot more covariates to adjust for, you may want to use some kind of penalized regression. For example, you could use the Lasso (L1-penalized regression); see Bloniarz et al. (2016).

Use out-of-sample predictions from any model

Maybe you instead want to use neural nets, trees, or an ensemble of a bunch of models? That’s fine, but if you want to be able to do valid statistical inference (i.e., get 95% confidence intervals that actually cover 95% of the time), you have to be careful. The easiest way to be careful in many Internet industry settings is just to use historical data to train the model and then get out-of-sample predictions Yhat from that model for your present experiment. You then then just subtract Yhat from Y and use the simple difference-in-means estimator. Aronow and Middleton (2013) provide some technical details and extensions. A simple extension that makes this more robust to changes over time is to use this out-of-sample Yhat as a covariate, as described above.3

  1. As Winston Lin notes in the comments and as is implicit in my comparison with post-stratification, as long as the number of covariates is small and not growing with sample size, the same asymptotic results apply. []
  2. Note that if the covariate is binary or, more generally, categorical, then this exactly coincides with the post-stratified estimator considered above. []
  3. I added this sentence in response to Winston Lin’s comment. []

Adjusting biased samples

Nate Cohn at The New York Times reports on how one 19-year-old black man is having an outsized impact on the USC/LAT panel’s estimates of support for Clinton in the U.S. presidential election. It happens that the sample doesn’t have enough other people with similar demographics and voting history (covariates) to this panelist, so he is getting a large weight in computing the overall averages for the populations of interest, such as likely voters:

There is a 19-year-old black man in Illinois who has no idea of the role he is playing in this election.

He is sure he is going to vote for Donald J. Trump.

And he has been held up as proof by conservatives — including outlets like Breitbart News and The New York Post — that Mr. Trump is excelling among black voters. He has even played a modest role in shifting entire polling aggregates, like the Real Clear Politics average, toward Mr. Trump.

As usual, Andrew Gelman suggests that the solution to this problem is a technique he calls “Mr. P” (multilevel regression and post-stratification). I wanted to comment on some practical tradeoffs among common methods. Maybe these are useful notes, which can be read alongside another nice piece by Nate Cohn on how different adjustment methods can yield very different polling results.


Complete post-stratification is when you compute the mean outcome (e.g., support for Clinton) for each stratum of people, such as 18-24-year-old black men, defined by the covariates X. Then you combine these weighting by the size of each group in the population of interest. This really only works when you have a lot of data compared with the number of strata — and the number of strata grows very fast in the number of covariates you want to adjust for.

Modeling sample inclusion and weighting

When people talk about survey weighting, often what they mean is weighting by inverse of the estimated probability of inclusion in the sample. You model selection into the survey S using, e.g., logistic regression on the covariates X and some interactions. This can be done with regularization (i.e., priors, shrinkage) since many of the terms in the model might be estimated with very few observations. Especially without enough regularization, this can result in very large weights when you don’t have enough of some particular type in your sample.

Modeling the outcome and integrating

You fit a model predicting the response (e.g., support for Clinton) Y with the covariates X. You regularize this model in some way so that the estimate for each person is going to “borrow strength” from other people with similar Xs. So now you have a fitted responses Yhat for each unique X. To get an estimate for a particular population of interest, integrate out over the distribution of X in that population. Gelman’s preferred version “Mr. P” uses a multilevel (aka hierarchical Bayes, random effects) model for the outcome, but other regularization methods may often be appealing.

This is nice because there can be some substantial efficiency gains (i.e. more precision) by making use of the outcome information. But there are also some practical issues. First, you need a model for each outcome in your analysis, rather than just having weights you could use for all outcomes and all recodings of outcomes. Second, the implicit weights that this process puts on each observation can vary from outcome to outcome — or even for different codings (i.e. a dichotomization of answers on a numeric scale) of the same outcome. In a reply to his post, Gelman notes that you would need a different model for each outcome, but that some joint model for all outcomes would be ideal. Of course, the latter joint modeling approach, while appealing in some ways (many statisticians love having one model that subsumes everything…) means that adding a new outcome to analysis would change all prior results.


Side note: Other methods, not described here, also work towards the aim of matching characteristics of the population distribution (e.g., iterative proportional fitting / raking). They strike me as overly specialized and not easy to adapt and extend.

Interpreting discrete-choice models

Are individuals random-utility maximizers? Or do individuals have private knowledge of shocks to their utility?

“McFadden (1974) observed that the logit, probit, and similar discrete-choice models have two interpretations. The first interpretation is that of individual random utility. A decisionmaker draws a utility function at random to evaluate a choice situation. The distribution of choices then reflects the distribution of utility, which is the object of econometric investigation. The second interpretation is that of a population of decision makers. Each individual in the population has a deterministic utility function. The distribution of choices in the population reflects the population distribution of preferences. … One interpretation of this game theoretic approach is that the econometrician confronts a population of random-utility maximizers whose decisions are coupled. These models extend the notion of Nash equilibrium to random- utility choice. The other interpretation views an individual’s shock as known to the individual but not to others in the population (or to the econometrician). In this interpretation, the Brock-Durlauf model is a Bayes-Nash equilibrium of a game with independent types, where the type of individual i is the pair (x_i, e_i). Information is such that the first component of each player i’s type is common knowledge, while the second is known only to player i.” — Blume, Brock, Durlauf & Ioannides. 2011. Identification of Social Interactions. Handbook of Social Economics, Volume 1B.

A deluge of experiments

The Atlantic reports on the data deluge and its value for innovation.1 I particularly liked how Erik Brynjolfsson and Andrew McAfee, who wrote the Atlantic piece, highlight the value of experimentation for addressing causal questions — and that many of the questions we care about are causal.2

In writing about experimentation, they report that Hal Varian, Google’s Chief Economist, estimates that Google runs “100-200 experiments on any given day”. This struck me as incredibly low! I would have guessed more like 10,000 or maybe more like 100,000.

The trick of course is how one individuates experiments. Say Google has an automatic procedure whereby each ad has a (small) random set of users who are prevented from seeing it and are shown the next best ad instead. Is this one giant experiment? Or one experiment for each ad?

This is a bit of a silly question.3

But when most people — even statisticians and scientists — think of an experiment in this context, they think of something like Google or Amazon making a particular button bigger. (Maybe somebody thought making that button bigger would improve a particular metric.) They likely don’t think of automatically generating an experiment for every button, such that a random sample see that particular button slightly bigger. It’s these latter kinds of procedures that lead to thinking about tens of thousands of experiments.

That’s the real deluge of experiments.

  1. I don’t know that I would call much of it ‘innovation’. There is some outright innovation, but a lot of that is in the general strategies for using the data. There is much more gained in minor tweaking and optimization of products and services. []
  2. Perhaps they even overstate the power of simple experiments. For example, they do not mention the fact that many times the results these kinds of experiments often change over time, so that what you learned 2 months ago is no longer true. []
  3. Note that two single-factor experiments over the same population with independent random assignment can be regarded as a single experiment with two factors. []

Applying social psychology

Some reflections on how “quantitative” social psychology is and how this matters for its application to design and decision-making — especially in industries touched by the Internet.

In many ways, contemporary social psychology is dogmatically quantitative. Investigators run experiments, measure quantitative outcomes (even coding free responses to make them amenable to analysis), and use statistics to characterize the collected data. On the other hand, social psychology’s processes of stating and integrating its conclusions remain largely qualitative. Many hypotheses in social psychology state that some factor affects a process or outcome in one direction (i.e., “call” either beta > 0 or beta < 0). Reviews of research in social psychology often start with a simple effect and then note how many other variables moderate this effect. This is all quite fitting with the dominance of null-hypothesis significance testing (NHST) in much of psychology: rather than producing point estimates or confidence intervals for causal effects, it is enough to simply see how likely the observed data is given there there is no effect.1 Of course, there have been many efforts to change this. Many journals require reporting effect sizes. This is a good thing, but these effect sizes are rarely predicted by social psychological theory. Rather, they are reported to aid judgments of whether a finding is not only statistically significant but substantively or practically significant, and the theory predicts the direction of the effect. Not only is this process of reporting and combining results not quantitative in many ways, but it requires substantial inference from the particular settings of conducted experiments to the present settings. This actually helps to make sense of the practices described above: many social psychology experiments are conducted in conditions and with populations that are so different from those in which people would like to apply the resulting theories, that expecting consistency of effect sizes is implausible.2 This is not to say that these studies cannot tell us a good deal about how people will behave in many circumstances. It's just that figuring out what they predict and whether these predictions are reliable is a very messy, qualitative process. Thus, when it comes to making decisions -- about a policy, intervention, or service -- based on social-psychological research, this process is largely qualitative. Decision-makers can ask, which effects are in play? What is their direction? With interventions and measurement that are very likely different from the present case, how large were the effects?3

Sometimes this is the best that social science can provide. And such answers can be quite useful in design. The results of psychology experiments can often be very effective when used generatively. For example, designers can use taxonomies of persuasive strategies to dream up some ways of producing desired behavior change.

Nonetheless, I think all this can be contrasted with some alternative practices that are both more quantitative and require less of this uneasy generalization. First, social scientists can give much more attention to point estimates of parameters. While not without its (other) flaws, the economics literature on financial returns to education has aimed to provide, criticize, and refine estimates of just how much wages increase (on average) with more education.4

Second, researchers can avoid much of the messiest kinds of generalization altogether. Within the Internet industry, product optimization experiments are ubiquitous. Google, Yahoo, Facebook, Microsoft, and many others are running hundreds to thousands of simultaneous experiments with parts of their services. This greatly simplifies generalization: the exact intervention under consideration has just been tried with a random sample from the very population it will be applied to. If someone wants to tweak the intervention, just try it again before launching. This process still involves human judgment about how to react to these results.5 An even more extreme alternative is when machine learning is used to fine-tune, e.g., recommendations without direct involvement (or understanding) by humans.

So am I saying that social psychology — at least as an enterprise that is useful to designers and decision-makers — is going to be replaced by simple “bake-off” experiments and machine learning? Not quite. Unlike product managers at Google, many decision-makers don’t have the ability to cheaply test a proposed intervention on their population of interest.6 Even at Google, many changes (or new products) under consideration are too difficult to build to them all: one has to decide among an overabundance of options before the most directly applicable data could be available. This is consistent with my note above that social-psychological findings can make excellent inspiration during idea generation and early evaluation.

  1. To parrot Andrew Gelman, in social phenomena, everything affects everything else. There are no betas that are exactly zero. []
  2. It's also often implausible that the direction of the effect must be preserved. []
  3. Major figures in social psychology, such as Lee Ross, have worked on trying to better anticipate the effects of social interventions from theory. It isn’t easy. []
  4. The diversity of the manipulations used by social psychologists ostensibly studying the same thing can make this more difficult. []
  5. Generalization is not avoided. In particular, decision-makers often have to consider what would happen if an intervention tested with 1% of the population is launched for the whole population. There are all kinds of issues relating to peer influence, network effects, congestion, etc., here that don’t allow for simple extrapolation from the treatment effects identified by the experiment. Nonetheless, these challenges obviously apply to most research that aims to predict the effects of causes. []
  6. However, Internet services play a more and more central role in many parts of our life, so this doesn’t just have to be limited to the Internet industry itself. []