Optimized Covariance Design for AB Test on Social Network under Interference

Online A/B tests have become increasingly popular and important for social platforms. However, accurately estimating the global average treatment effect (GATE) has proven to be challenging due to network interference, which violates the Stable Unit Treatment Value Assumption (SUTVA) and poses a great challenge to experimental design. Existing network experimental design research was mostly based on the unbiased Horvitz-Thompson (HT) estimator with substantial data trimming to ensure unbiasedness at the price of high resultant estimation variance. In this paper, we strive to balance the bias and variance in designing randomized network experiments. Under a potential outcome model with 1-hop interference, we derive the bias and variance of the standard HT estimator and reveal their relation to the network topological structure and the covariance of the treatment assignment vector. We then propose to formulate the experimental design problem to optimize the covariance matrix of the treatment assignment vector to achieve the bias and variance balance by minimizing a well-crafted upper bound of the mean squared error (MSE) of the estimator, which allows us to decouple the unknown interference effect component and the experimental design component. An efficient projected gradient descent algorithm is presented to implement the desired randomization scheme. Finally, we carry out extensive simulation studies 2 to demonstrate the advantages of our proposed method over other existing methods in many settings, with different levels of model misspecification.

Paper

References (43)

Scroll for more · 31 remaining

Similar papers

Peer review

Reviewer EFcj6/10 · confidence 3/52023-07-03

Summary

This paper focused on designing a randomization scheme at the cluster level for the A/B test. It proposed and derived an upper bound for the MSE of the HT estimator, which was targeted to be minimized to optimize the experiment design. This article treated the covariance matrix of the treatment vector as the decision variable in optimization to represent a class of design. Besides, it proposed to adopt projected gradient descent to guarantee the validity of the optimized covariance matrix in order to generate a randomization scheme that supports legitimate sampling. Also, it stressed the significance of boas and attempted to balance bias and variance better. Systematic simulations on a real-time social network were conducted. Experiment outcomes were analyzed and different methods were compared to show the effectiveness of the newly proposed method in this article.

Strengths

1. This paper is generally easy to follow. 2. The problem is well formulated in a new perspective, with detailed analysis and deduction. 3. The effectiveness of the proposed method is verified in the experiment. Horizontal comparisons with existing methods are sufficient and effective.

Weaknesses

1. Insufficient background introduction. Different concepts in the area should be introduced more specifically and it is better to point out their connections. 2. Unclear notation definitions and explanations, leading to confusion in understanding. 3. Experiments were not comprehensive, or at least not comprehensively expressed.

Questions

1. Where is the definition of di that appeared below line 150 in assumption 1? (Does it refer to the degree of node i?) It is hard to read when definitions are missed or distributed in different parts. It is better to summarize all of them into one table with clear definitions. 2. In section 2.1, line 132, what are the certain community detection algorithms? It is better to list out some works with references. 3. The experiment part seemed to be written in a hurry. Even though necessary comparison methods were involved, many other essential details such as the basic information about the dataset (i.e., edges, nodes, degrees, etc.) were missing. Also, it seems that experiments were only implemented on one dataset, which made the experiment outcome not convincing enough. 4. What is the real-world application of this method? Please offer some direct applications such as the areas that may benefit from your method. This is important since your derivation used different assumptions when dealing with difficult parts in the optimization. Will the assumptions affect its performance in general? Is the proposed method only workable under a specific scene and what is the case?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

2 fair

Presentation

3 good

Contribution

2 fair

Limitations

Limitations of this article were clearly expressed in this article. For instance, the upper bound should be tighter in the optimization. And different other cases like the non-balanced treatment assignment have not been considered yet. The basic setting beyond nationhood interference assumption warrants a further extension. Besides, assumptions may hinder the method from applying to a more generalized case.

Reviewer 1xr36/10 · confidence 5/52023-07-04

Summary

In this paper, authors present a novel experimental design to be used for randomized experiments under network interference. Authors begin by considering a baseline adjusted Horvitz--Thompson estimator (which crucially requires knowledge of $Y_i(\mathbf{0})$) and a pre-specified clusterings. Given these two things, authors derive a bound on the mean squared error of the estimator under an arbitrary cluster design. Motivated by a Grothendeick identity, authors propose an experimental design where a normal vector is sampled from a fixed covariance matrix $\Sigma$ and the signs of the vector become treatment assignment. Authors propose to select an experimental design by (essentially) using the normal covariance matrix $\Sigma$ as a decision variable and the upper bound on the mean squared error as an objective. Finally, simulations are run to investigate the performance of the method under various types of model-mispecification.

Strengths

The main strength of the paper is a novel method for designing randomized experiments under network interference. The novelty of the method comes from developing an objective function whose decision variables are continuous, (essentially) representing the covariance matrix of assignments. This stands in contrast to previous methods which focus on selecting clusters for independent cluster randomization designs. An additional strength of the paper is that the objective considers both bias and variance, which is uncommon: most approaches attempt to "fix" one of these two things, but typically not both. One of the most exciting technical connections is the use of Grothendieck's identity, which (to the best of my knowledge) has not been used in causal inference. This offers a new tool in the design of experiments, which I suspect will be useful beyond the specific design used in this paper. Moreover, the simulations are very well formulated and executed. In particular, the authors investigate the effectiveness of their design under various forms of model misspecification, which speaks to the robustness of the design. This is important to investigate via simulations, as formal theory seems difficult given the black-box nature of the experimental design.

Weaknesses

There are several weaknesses of the current method. 1. **Assumed Knowledge**: The paper makes some strong assumptions as to what the experimenter knows. For example, authors assume that the coefficients $\alpha_i$ are known by the experimenter. I think most experimenters will find that knowledge of each individual $\alpha_i$ is too strong of an assumption to be practical -- in the SUTVA setting, this implies that all individual treatment effects can be estimated perfectly without any randomization. In order to be transparent, authors should state this assumption and estimator earlier in the paper, perhaps replacing the HT estimator in (eq 5). 2. **Pre-specified Clusters**: The paper assumes that clusters are pre-specified and little advice is given to practicioners on how to select the clusters in order to minimize MSE. 3. **Understanding of the Variance**: In order to run power calculations, experimenters should have at least a rough understanding (e.g. asymptotic rates) of how the variance of the experimental design depends on sample size $n$ and network parameters. Because this method is based on a black-box optimization procedure, it seems hard to analyze the variance (i.e. optimal value). 4. **Confidence Intervals**: In practice, experimenters value interval estimators (i.e. confidence intervals) more than point estimators, as they provide a method for uncertainty quantification. The necessary tools for uncertainty quantification (e.g. Central Limit Theorem, variance estimation) are not presented in this paper. While authors should make certain minor changes to address the above, I think that these weaknesses actually constitute further research directions on this exciting method. Overall, it is my opinion that the strengths of the paper outweigh the weaknesses. In the two sections below, I discuss minor weakenesses in the technical discussions and literature review which should be addressed by authors before publication. ## Technical Remarks Below are some minor remarks on technical aspects of the paper. I think there are a few technical issues in the discussions, but I believe these can be easily fixed and will strengthen the technical contribution of the paper. 1. (Line 137) authors write "without loss of generaltiy, we consider the balanced cluster-level randomization scheme satisfying...$E[z_i] = 1/2$." I think that this restriction is perfectly fine, but I would say it is not technically correct to describe it as "without loss of generality". Consider the usual SUTVA setting: if the outcomes under treatment have more variation than outcomes under control, then the Horvitz--Thompson estimator can be made to have smaller variance by setting $p$ so that treatment is assigned more frequently. I think the phrase "without loss of generality" is not warranted and a simple fix is to just remove it. 2. (Line 152) The word "overparametrized" is not quite standard in this literature so I'd recommend either informally defining it, or just saying that "there are more unknown potential outcomes than observations". This is true even under SUTVA. 3. (Line 208): Authors introduce what they call "Assumption 2". I would refer to this as a "condition" rather than an assumption. The reason is that Assumption 2 only plays a role in choosing the design -- using standard techniques (CLT + variance estimator), Assumption 2 would not be necessary for, say, the validity of confidence intervals. Some readers might misinterpret the use of the word "Assumption" to think "if this condition does not hold, then the estimates are no longer statistically valid in some sense". 4. (Line 254) Authors write "This lemma enables us to sample from bivariates Bernoulli distribution with mean (1/2, 1/2) and any valid covariance". While this is true for $n=2$ variables, I do not believe it to be true for general $n$ variables. It is sort of implied in the Section that this "Grotendeick mapping" can recover any covariance matrix of $\pm 1$ variables. If authors have a proof of this, they should provide it; otherwise, they should clearly state that this "Groethendeick mapping" cannot generate all $\pm 1$ covariance matrices. I think that clarifying this will increase understanding and appreciation of the method. 5. (Line 294): The Monte Carlo simulation is only performed 200 times. In my experience, this is quite low. Can you increase to 10,000 Monte Carlo runs before camera ready submission? This will increase the reader's confidence in you results. ## Literature Review + References The authors have missed a few important references in the causal inference literature. I believe these should be easily fixable and would increase the relevancy of the paper by better situating it within the causal inference literature. 1. (Line 36-37) In reference to cluster designs, authors write: "This technique is originally developed in [31] and becomes a prevalent paradigm for network experiment design". I completely agree that [31] was an influential paper for introducing cluster designs to the computer science community. [35] works the exposure mapping framework for interference [1] which allows for this arbitrary network interfernece. However, the so-called "partial interference" assumption has been used since at least Hudgens & Halloran (2008) and cluster designs were advocated for here. So, I'd at least reference some of this early work on clustering in the context of partial interference. I think this will also tie your contributions back to the vien of causal inference in the statistics literature in a stronger way. 2. (Line 38) Authors write that "sharing same treatment within cluster is usually necessary for characterizing the GATE". I would remove or substantially weaken this statement. The proliferation of cluster designs is *not* because they are necessary; but rather, a conceptually simple yet effective type of experimental design. 3. (Line 64) Authors write "In this paper, we propose to treat the covariance matrix of treatment vector as a decision variable in optimization". The following paper seems especially relevant and authors should draw a comparison: Harshaw et al (2019) "Balancing Covariates in Randomized Experiments using the Gram--Schmdit Walk Design". This paper studies experimental designs that directly control the covariance matrix Cov(z) (via discrepancy theory) to bound the variance of Horvitz--Thompson estimator by an implicit ridge regression of outcomes on covariates. A key idea in that paper is to use the operator norm as a measure of worst-case variance, which seems like an alternative to your Assumption 2. Given the similarity in the spirit of the two papers, this paper would benefit from a brief comparison discussion. 4. (Line 94) Authors cite several papers on bipartite experiments. It seems that [15] and [16] are duplicates. To the best of my knowledge, the paper of Zigler & Papadogeorgou (2021) was the first paper to propose bipartite experiments (it has been a working paper since 2019), so a citation is warranted in that discussion. 5. (Line 104): Authors write "Along the same direction, [32] tries to provide..." I recommend revising this language. The word "tries" gives the indication that "[32] tries and fails". 6. (Line 108): In the discussion of partial interference, early work like Hudgens and Halloran (2008) is missing. 7. (Line 116): A citation of several papers with more general forms of interference is listed. The recent paper Harshaw, Sävje, Wang (2022) "A design-based riesz representation framework for randomized experiments" is worth citing, as it proposes a deisgn-based framework which captures and extends previous types of interference. 8. (Line 251): I haven't read [19], but it was my understanding that Grothendieck's identity is typically a different type of statement, where there are two fixed vectors $x$ and $y$ and the random variables are $\textrm{sign}(\langle z , x \rangle)$ and $\textrm{sign}(\langle z , y \rangle)$, where $z$ is uniform from the $\ell_2$ ball. In fact, I have only seen Lemma 1 in certain course notes on Sums-of-Squares (though I'm sure it's appeared in other places). If Lemma 1 does not directly appear in [19] then authors should cite a relevant paper that derives it. In fact, this would probably be helpful to tie your work back to theoretical computer science's use of the technique. ## References - Harshaw, C., Sävje, F., Spielman, D., & Zhang, P. (2019). "Balancing covariates in randomized experiments with the Gram-Schmidt Walk design". (arXiv:1911.03071) - Harshaw, C., Sävje, F., & Wang, Y. (2022). "A design-based riesz representation framework for randomized experiments". (arXiv:2210.08698) - Hudgens, M. G., & Halloran, M. E. (2008). "Toward causal inference with interference". Journal of the American Statistical Association, 103(482), 832–842. - Zigler, C. M. and Papadogeorgou, G. (2021). "Bipartite causal inference with interference". Statist. Sci., 36(1):109–123.

Questions

1. One way to avoid "picking clusters" is to run the method when each unit is its own singleton cluster. How does the proposed method perform in this case? In other words, is the specification of clusters necessary to achieve small variance? If so, can you describe why or generally comment more on this? (this is a discussion I would find really fascinating in the paper) 2. I claimed that the "Grothendeick mapping" cannot recover all covariance matrices on $\pm 1$ matrices. Can you comment on this? 3. Can you increase the Monte Carlo samples to 10,000? 4. Can you comment on the comparison to the Gram--Schmidt Walk Design of Harshaw et al (2019)?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

5: You are absolutely certain about your assessment. You are very familiar with the related work and checked the math/other details carefully.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

yes

Reviewer Q5Zi7/10 · confidence 3/52023-07-06

Summary

In this paper, the authors propose a new algorithm for A/B test design under network effect with cluster level randomization. By derivation of an upper MSE upper bound with bias-variance trade-off, the authors reparameterize to directly optimize the covariance of the treatment vector. An efficient PGD algorithm is proposed for efficient optimization. Simulation study on a real-world network demonstrates the effectiveness of the proposed method.

Strengths

1. The authors study a very important and practical problem in A/B test design with reasonable assumptions on cluster level randomization. 2. The paper is well written and the proposed method is both explained clearly, with reasonable assumptions and well justified with detailed derivations. 3. The proposed reparameterization and the PGD optimization algorithm makes the algorithm practical and easy to implement.

Weaknesses

1. It seems that the clustering algorithm plays a very important role in the quality of the A/B experiment design. From the covariate matrix formulation, maybe in some cases it would be better to directly merge two clusters instead of using the treatment vector to guarantee the same treatment. It would be very interesting to see how the proposed method performs under different cluster methods /parameters besides the number of clusters. 2. As A/B experiments are usually used in real large-scale networks, it would be interesting to see scalability results in terms of number of clusters for different methods under comparison. 3. As the authors point out in the future work, it would also be interesting to see some empirical evaluation on the tightness of the proposed upper bound on the MSE. It may provide some insight on whether further tightening the bound can bring additional benefits.

Questions

Please see above

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

Yes

Reviewer 2fwK5/10 · confidence 3/52023-07-24

Summary

This paper presents an optimized covariance design for A/B tests on social networks with interference. The authors address the challenge of accurately estimating the global average treatment effect (GATE) in the presence of network interference. They propose a method to balance bias and variance in experimental design by optimizing the covariance matrix of the treatment assignment vector. The paper derives the bias and variance of the estimator and proposes an algorithm to implement the desired randomization scheme. Simulation studies demonstrate the advantages of their method over existing methods in various settings. Overall, the paper contributes an approach to improve the estimation of treatment effects in social network experiments by optimizing the covariance design.

Strengths

This paper presents a approach to optimize the covariance design for A/B tests on social networks with interference, which improves the estimation of treatment effects and enables more accurate decision making based on experimental results. The paper is well-written and clearly explains the problem formulation, assumptions, and the proposed method. The authors provide a thorough analysis of the bias and variance of the estimator, a upper bound for the MSE of the HT estimator, and a projected gradient descent algorithm to solve the optimization problem. The simulation studies demonstrate the effectiveness and robustness of the proposed method in various settings.

Weaknesses

1. The paper briefly mentions assumptions about the direct treatment effect and interference effect. However, a more thorough discussion on the validity and generalizability of these assumptions would be valuable. Additionally, discussing the limitations of the proposed method, such as its sensitivity to certain parameter values or potential biases introduced by the assumptions, would provide a more comprehensive understanding of the method's applicability and potential drawbacks. 2. In the experimental results presented in Tables 2, 5, 6, 7, 10, 11, 14, 15, 19, 22, and 23 in the appendix, it is observed that the proposed algorithm’s performance in terms of MSE is significantly weaker compared to other benchmark algorithms such as ReAR, IBR-p, and PSR. This suggests that further improvements may be necessary for the proposed algorithm to achieve competitive performance 3. The paper could benefit from a more detailed introduction and explanation of the motivation behind balancing bias and variance. Providing additional context and background information on this topic would help to strengthen the overall argument and improve the clarity of the paper.

Questions

Please refer to weaknesses 1, 2, and 3

Rating

5: Borderline accept: Technically solid paper where reasons to accept outweigh reasons to reject, e.g., limited evaluation. Please use sparingly.

Confidence

3: You are fairly confident in your assessment. It is possible that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work. Math/other details were not carefully checked.

Soundness

3 good

Presentation

3 good

Contribution

2 fair

Limitations

While the paper briefly mentions assumptions about the direct treatment effect and interference effect, a more thorough discussion on the validity and generalizability of these assumptions would be valuable. Additionally, it would be beneficial for the authors to include a discussion on the limitations and potential negative consequences of their work. This could include an examination of the proposed method’s sensitivity to certain parameter values or potential biases introduced by the assumptions. Providing a more comprehensive understanding of the method’s applicability and potential drawbacks would strengthen the overall manuscript

Reviewer 1xr32023-08-11

response to authors

I thank the authors for their thoughtful response to my review. I think many of my questions were answered appropriately. I want to just emphasize a few aspects of my review: - It seems that we are both in agreement that the Grothendeick mapping cannot generate all +/- 1 covariance matrices. All I recommend here is that the language is updated so that readers do not get this impression. - I think your paper makes a good contribution without the development of confidence intervals. I raised this point only as a weakness, not one that I think needs to be addressed in this paper.

Authorsrebuttal2023-08-13

Thank you once more for your dedicated efforts in providing feedback on our paper. We will certainly add comments after this identity to enhance the clarity. Additionally, the issues remarked as weaknesses will be discussed in the future directions of the upcoming version.

Reviewer Q5Zi2023-08-13

Thank you for the rebuttal

Thanks for the response. I have read the rebuttal and remain my score.

Authorsrebuttal2023-08-13

Thank you once more for your decision and efforts in the review process, and we'll supplement discussion on your questions in the upcoming version.

Authorsrebuttal2023-08-13

Thank you once more for your decision and the time you've invested in reviewing our paper, and we will supplement discussion on your concerns in the upcoming version for enhancing the positioning of this paper.

Reviewer EFcj2023-08-18

Thanks for your reply.

I have carefully read your response. The supplementary experiments complete the work and it solves of my previous doubts. I believe this work will contribute to the community and would like to raise my rating to 6.

Authorsrebuttal2023-08-19

Thank you again for your approval on this paper. We'll present the simulation result in a clearer way in the upcoming version.

Program Chairsdecision2023-09-21

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC