There are two general views in causal analysis of experimental data: the super population view that the units are an independent sample from some hypothetical infinite population, and the finite population view that the potential outcomes of the experimental units are fixed and the randomness comes solely from the treatment assignment. These two views differs conceptually and mathematically, resulting in different sampling variances of the usual difference-in-means estimator of the average causal effect. Practically, however, these two views result in identical variance estimators. By recalling a variance decomposition and exploiting a completeness-type argument, we establish a connection between these two views in completely randomized experiments. This alternative formulation could serve as a template for bridging finite and super population causal inference in other scenarios.
Neyman [1, 2] defined causal effects in terms of potential outcomes, and proposed an inferential framework viewing all potential outcomes of a finite population as fixed and the treatment assignment as the only source of randomness. This finite population view allows for easy interpretation free of any hypothetical data generating process of the outcomes, and is used in a variety of contexts e.g., [3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22]. This approach is considered desirable because, in particular, it does not assume the data are somehow a representative sample of some larger (usually infinite) population.
Alternative approaches, also using the potential outcomes framework, assume that the potential outcomes are independent and identical draws from a hypothetical infinite population. Mathematical derivations under this approach are generally simpler, but the approach itself can be criticized because of this typically untenable sampling assumption. Furthermore, this approach appears to ignore the treatment assignment mechanism.
That being said, it is well known that the final variance formulae from either approach tend to be quite similar. For example, while the variance of the difference-in-means estimator for a treatment-control experiment under an infinite population model is different from the one under Neyman’s  finite population formulation, this difference is easily represented as a function of the variance of the individual causal effects. Furthermore, this difference term is unidentifiable and is often assumed away under a constant causal effect model ([1, 2, 23, 24, 25, 26]), or by appeals to the final estimators being “conservative.”
For the difference in means, the infinite population variance estimate gives a conservative (overly large) estimate of the finite population variance. As deriving infinite population variance expressions, relative to finite population variance expressions, tends to be more mathematically straightforward, we might naturally wonder if we could use infinite population expressions as conservative forms of finite population expressions more generally. In this work we show that in fact we can assume an infinite population model as an assumption of convenience, and derive formula from this perspective. This shows that we can thus consider the resulting formula as focused on the treatment assignment mechanism and not on a hypothetical sampling mechanism, i.e., we show variance derivations under the infinite population framework can be used as conservative estimators in a finite context.
Mathematically, this result comes from a variance decomposition and a completeness-style argument characterizing the connection and the difference between these two views. The variance decomposition we use has previously appeared in Imai , Imbens and Rubin , and Balzer et al. . The completeness-style argument, which we believe is novel in this domain, then sharpens the variance decomposition by moving from an expression on an overall average relationship to one that holds for any specific sample.
Our overall goal is simple: we wish to demonstrate that if one uses variance formula derived from assuming an infinite population sampling model, then the resulting inference one obtains will be correct with regards to the analogous sample-specific treatment effects (although it could be potentially conservative in that the standard errors may be overly large) regardless of the existence of any sampling mechanism.
2 Super population, finite population, and samples
Assume that random variables represent the pair of potential outcomes of an infinite super population, from which we take an independent and identically distributed (IID) finite population of size :
The individual causal effect for unit is . We first discuss completely randomized experiments, and comment on other experiments in Section 4. For a completely randomized experiment, we randomly assign units to receive treatment, leaving the remaining units to receive control. Let be the treatment assignment vector, which takes a particular value with probability , for any with . The observed outcome for unit is then
At the super population level, the average potential outcomes are and , and the average causal effect is
The population variances of the potential outcomes and individual causal effect are
At the finite population level, i.e., for a fixed sample , the average potential outcomes and average causal effect are
The corresponding finite population variances of the potential outcomes and individual causal effects are
where, following the tradition of survey sampling , we use the divisor . \;All these quantities are fully dependent on . In classical casual inference , the potential outcomes of these experimental units, , are treated as fixed numbers. Equivalently, we can consider such causal inference to be conducted conditional on e.g., [5, 6, 18, 19, 20].
Regardless, we have two parameters – the population average treatment effect , and the sample average treatment effect . After collecting the data, we would want to draw inference about or .
Our primary statistics are the averages of the observed outcomes and the difference-in-means estimator:
We also observe the sample variances of the outcomes under treatment and control using
We do not have the sample analogue of or because and are never jointly observed for any unit in the sample.
We summarize the infinite population, finite population and sample quantities in Table 1.
3 Deriving complete randomization results with an independent sampling model
The three levels of quantities in Table 1 are connected via independent sampling and complete randomization. Neyman , without reference to any infinite population and by using the assignment mechanism as the only source of randomness, represented the assignment mechanism via an urn model, and found
He then observed that the final term was unidentifiable but nonnegative, and thus if we dropped it we would obtain an upper bound of the estimator’s uncertainty.
We next derive this result by assuming a hypothetical sampling mechanism from some assumed infinite super-population model of convenience. This alternative derivation of the above result, which can be extended to other assignment mechanisms, shows how we can interpret formulae based on super-population derivations as conservative formulae for finite-sample inference.
3.1 Sampling and randomization
To begin, note that IID sampling of from the super population implies three things: first, the finite population average causal effect satisfies , second
and third, the sample variances are unbiased estimates of the true variances:
Conditional on , randomization of the treatment is the only source of randomness. In a completely randomized experiment, the outcomes in the treatment group form a simple random sample of size from , and the outcomes in the control group form a simple random sample of size from . Therefore, classical survey sampling theory  for the sample mean and variance gives
We do not use the notation , because, depending on context, such notation could either indicate expectation conditional on or expectation averaged over . We therefore use conditional expectations and conditional variances explicitly.
If we do not condition on , then the independence induced by the assignment mechanism means the outcomes under treatment are IID samples of and the outcomes under control are IID samples of , and furthermore these samples are independent of each other. This independence makes it straightforward to show that is unbiased for with super population variance
This is the classic infinite population variance formula for the two sample difference-in-means statistic. We could use it to obtain standard errors by plugging in and for the two variances.
3.2 Connecting the finite and infinite population inference with a variance decomposition
We will now extend the above to indirectly derive the result on the variance of for finite population inference without explicitly enumerating the potential outcomes. The variance decomposition formula implies
which further implies that the finite population variance of satisfies (using eq. (3))
Compare to the classic variance expression (1), which is this without the expectation. Here we have that on average our classic variance expression holds. Now, because this is true for any infinite population, as it is purely a consequence of the IID sampling mechanism and complete randomization, we can close the gap between eqs (1) and (8). Informally speaking, because eq. (8) holds as an average over many hypothetical super populations, it should also hold for any finite population at hand, and indeed it does, as we next show using a “completeness” concept from statistics .
3.3 A “Completeness” argument
a function of a fixed finite sample , as the difference of the hypothesized finite sample variance formula and the actual finite sample variance. Formula (8) shows . Now we are going to show the stronger result .
For any given sample , we have fixed sample quantities: and Some algebra gives
for some , and . is the same regardless of the ordering of units, so by symmetry the constants , and must be the same regardless of and . This gives for all . Thus
Because , we have
Similarly, because , we have
Use the above to replace our cross terms of with single index terms to write as
where and . Because , we have where
Because eq. (11) holds for any populations regardless of its values of and , it must be true that which implies .
Equation (8) relies on the assumption that the hypothetical infinite population exists, but eq. (1) does not. However, the completeness-style argument allowed us to make our sampling assumption only for convenience in order to prove eq. (1) by, in effect, dropping the expectation on both sides of eq. (8). Similar argument exists in the classical statistics literature; see Efron and Morris  for the empirical Bayes view of Stein’s estimator. While the final result is, of course, not new, we offer it as it gives an alternative derivation that does not rely on asymptotics such as a growing super population or a focus on the properties of the treatment assignment mechanism.
Using Freedman’s  results, Aronow et al.  considered a super population with units, with the finite population being a simple random sample of size . Letting , we can obtain similar results. We go in the other direction: we use the variance decomposition of eq. (6) to derive the finite population variance from the super population one.
This decomposition approach also holds for other types of experiments. First, for a stratified experiment, each stratum is essentially a completely randomized experiment. Apply the result to each stratum, and then average over all strata to obtain results for a stratified experiment. Second, because a matched-pair experiment is a special case of a stratified experiment with two units within each stratum, we can derive the Neyman-type variance cf. ([7, 18]) directly from that of a stratified experiment. Third, a cluster-randomized experiment is a completely randomized experiment on the clusters. If the causal parameters can be expressed as cluster-level outcomes, then the result can be straightforwardly applied cf.([11, 17]). Fourth, for general experimental designs, the variance decomposition in eq. (6) still holds, and therefore we can modify the derivation of the finite population variance according to different forms of eqs (4) and (5).
In a completely randomized experiment, the finite population sampling variance of in eq. (1) depends on three terms: the first two can be unbiasedly estimated by and , but the third term is unidentifiable from the data. Assuming a constant causal effect model, and the variance estimators under both finite and super population inference coincide. However, Aronow et al. , Robins  and Ding and Dasgupta  demonstrated that the treatment variation term has a sharp lower bound that may be larger than , which allows for more precise variance estimators under the finite population view. This demonstrates that we can indeed make better inference conditional on the sample we have. On the other hand, in this work, we showed that assuming an infinite population, while not necessarily giving the tightest variance expressions, nonetheless gives valid (conservative) variance expressions from a finite-population perspective. We offer this approach as a possible method of proof that could ease derivations for more complex designs. More broadly, it is a step towards establishing that infinite population derivations for randomized experiments can be generally thought of as pertaining to their finite population analogs. Also see Lin  and Samii and Aronow , who provide alternative discussions of super population regression-based variance estimators under the finite population framework.
Our discussion is based on the frequentists’ repeated sampling evaluations of the difference-in-means estimator for the average causal effect. In contrast, Fisher  proposed the randomization test against the sharp null hypothesis that for all units, which is numerically the same as the permutation test for exchangeable units sampled from an infinite population . This connection becomes more apparent when only the ranks of the outcomes are used to construct the test statistics, as discussed extensively by Lehmann .
We thank Dr. Peter Aronow (the Associate Editor) and three anonymous reviewers for helpful comments.
1. Neyman J. On the application of probability theory to agricultural experiments. essay on principles (with discussion). Section 9 (translated). Reprinted ed. Stat Sci 1923;5:465–72.Search in Google Scholar
2. Neyman J. Statistical problems in agricultural experimentation (with discussion). J Roy Stat Soc 1935;2:107–80.Search in Google Scholar
4. Hinkelmann K, Kempthorne O. Design and analysis of experiments, volume 1: introduction to experimental design, 2nd ed. New Jersey: John Wiley & Sons, Inc., 2008.Search in Google Scholar
5. Copas J. Randomization models for the matched and unmatched 2×2 tables. Biometrika 1973;60:467–76.Search in Google Scholar
14. Miratrix LW, Sekhon JS, Yu B. Adjusting treatment effect estimates by post-stratification in randomized experiments. J Roy Stat Soc Ser B (Stat Methodol) 2013;75:369–96.10.1111/j.1467-9868.2012.01048.xSearch in Google Scholar
18. Imbens GW, Rubin DB. Causal inference for statistics, social and biometrical sciences: an introduction. Cambridge: Cambridge University Press, 2015.10.1017/CBO9781139025751Search in Google Scholar
22. Li X, Ding P. General forms of finite population central limit theorems with applications to causal inference. J Am Stat Assoc 2017. (in press).10.1080/01621459.2017.1295865Search in Google Scholar
23. Hodges JL, Lehmann EL. Basic concepts of probability and statistics. San Francisco: Holden-Day, 1964.Search in Google Scholar
25. Reichardt CS, Gollob HF. Justifying the use and increasing the power of a t test for a randomized experiment with a convenience sample. Psychol Methods 1999;4:117–28.10.1037/1082-989X.4.1.117Search in Google Scholar
26. Freedman D, Pisani R, Purves R. Statistics, 4th ed. New York: W. W. Norton & Company, 2007.Search in Google Scholar
27. Balzer LB, Petersen ML, Laan MJ. Targeted estimation and inference for the sample average treatment effect in trials with and without pair-matching. Stat Med 2016;35:3717–32.10.1002/sim.6965Search in Google Scholar PubMed PubMed Central
28. Cochran WG. Sampling techniques, 3rd ed. New York: John Wiley & Sons, 1977.Search in Google Scholar
29. Lehmann EL, Romano JP. Testing statistical hypotheses, 3rd ed. New York : Wiley, 2008.Search in Google Scholar
30. Efron B, Morris C. Stein’s estimation rule and its competitors – an empirical Bayes approach. J Am Stat Assoc 1973;68:117–130.Search in Google Scholar
33. Samii C, Aronow PM. On equivalencies between design-based and regression-based variance estimators for randomized experiments. Stat Probab Lett 2012;82:365–70.10.1016/j.spl.2011.10.024Search in Google Scholar
34. Fisher RA. The design of experiments, 1st ed. Edinburgh, London: Oliver and Boyd, 1935.Search in Google Scholar
35. Lehmann EL. Nonparametrics: statistical methods based on ranks, 1st ed. San Francisco: Holden-Day, 1975.Search in Google Scholar
© 2017 Walter de Gruyter GmbH, Berlin/Boston
This article is distributed under the terms of the Creative Commons Attribution Non-Commercial License, which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.