Agency Fallibility and the Effect of Judicial Review in Competition Cases

Senior Managing Director, Jorge Padilla, recently authored an article for EU Law Live’s Competition Corner on the judicial review in competition cases and whether the appeal process reduces the cost of error.

The job of an administrative agency when assessing a competition complaint is to produce a decision {infringement, no infringement} by assessing the complainant’s theory of harm (or complainant’s narrative) and the defendant’s counterarguments (or defendant’s narrative) in light of the available evidence. In reaching a decision, because of human fallibility, the agency can make two types of error: it may conclude (i) that the complainant is right when it is not (a Type I error or a false positive), or (ii) that the defendant is right when it is not (a Type II error or a false negative). Each of these errors involves a cost – i.e., a loss of competition to the detriment of consumers. As Judge Easterbrook explained, if the agency chooses its decision to minimize the expected cost of error, that decision will likely be the one that maximizes consumer welfare by protecting the competitive process.

Under the expected error cost minimization principle, the agency’s decision should minimize the expected cost of error by finding that the defendant infringed the competition laws when the available evidence is more consistent with the complainant’s narrative than the defendant’s narrative. Otherwise, the error cost minimization principle would imply a finding of no infringement.

The difficulty is that, in many circumstances, the two narratives are vague and, therefore, it is complex, when not impossible, to use the available evidence to discriminate between them (i.e., discern which one is supported by the evidence). It is typically easier to determine which of the competing narratives – the complainant’s or the defendant’s – better explains the facts. Assessing which narrative fits the evidence better and, therefore, is relatively more plausible, is precisely what, according to Pardo and Allen (Juridical Proof and the Best Explanation, Law and Philosophy, 2008), courts do.

But how much better? What are the determinants of the threshold of persuasion that must be met to conclude that one narrative is more plausible than the other? The literature shows that this threshold depends on the relative costs of the Type I and Type II errors. The threshold of persuasion for a finding of infringement is lower when the cost of the Type II error is low relative to the cost of a Type I error. The threshold of persuasion is also influenced by the decision-maker’s “perception of normality”, anchored in prior experience and personal knowledge, and framed by the relevant rules and case law. In other words, the threshold of persuasion for a finding of infringement is lower when, based on experience, the theory of harm is a priori more plausible than the alternative pro-competitive motivation.

Yet, even decisions made using the expected error cost minimization principle may be erroneous, i.e. result in false positives or negatives. This is why society, it is often said, makes use of the judicial review, or appeals, process to further control the expected cost of error. Firstly, the costs of the appeals process are likely be smaller than the cost of a random review process (where all decisions whether right or wrong would be candidates for revision), since litigants are likely to seek reconsideration of cases where errors were probably made. Secondly, the appeals process may discipline lower-court judges or administrative agencies and increase trial or inquisitorial accuracy. Judges and agencies fear reversal because of its impact on their reputation and their prospects of promotion. Decision-makers, whether judges or civil servants working in competition authorities, seeking to avoid reversal are likely to expend more effort in their enquiries. They may also be less likely to adopt decisions which reflect their own ideological biases or the interests of the groups they favour, and focus more on protecting social efficiency and the interests of the collective body which they represent.

To investigate whether the judicial review process is likely to result in less costly errors, consider the following stylized model. Let x denote the net effect on consumer welfare of a business practice undertaken by a firm with market power. From the perspective of a competition agency, x is probabilistic: it can be positive (pro-competitive) or negative (anti-competitive). Suppose the probability that the agency concludes that the conduct is anticompetitive is p(x). If the agency’s test were perfect, all practices with a positive welfare impact (x > 0) would be found to be pro-competitive and all practices with a negative welfare impact (x < 0) would be found to be anticompetitive. Yet, as explained above, no agency is error free. In consequence, we expect that the agency will find pro-competitive practices to be illegal with some positive probability and vice versa.

Suppose the appeals process works as follows: once the agency has reached its decision, the appeals court reviews only those practices that are found to be anticompetitive and all infringement decisions are appealed. Let the probability that the appeals court finds that the practice be anticompetitive is q(x), then the ex-ante probability that the practice is found to be an infringement after the two-stage review process is p(x)q(x) – i.e., the probability that both instances find the conduct at issue anticompetitive. Since q(x) ≤ 1, we have that p(x)q(x) ≤ p(x). In other words, the introduction of the appeals process does not increase and may even reduce the incidence of Type I errors, while it does not reduce and may even increase the incidence of type II errors. If Type I errors are more costly than Type II errors, then this appeals process does not increase and may even reduce the expected cost of error. The opposite is true if Type II errors are relatively more costly.

These results can change when the appeals court reviews the agency’s decisions irrespective of whether they are infringement or non-infringement decisions and all decisions are appealed. In that case, the ex-ante probability that the practice is found infringing after the two-stage review process is p(x)q(x) + (1 – p(x))q(x) – i.e., the probability that both instances find the conduct at issue anticompetitive plus the probability that the agency considers it procompetitive but its decision is reversed by the appeals court. Since this sum equals q(x), then the introduction of this alternative appeals process will reduce the likelihood of a Type I error (and increase that of a Type II error) when q(x) < p(x). It follows that the expected cost of error will not be changed by the appeals process if q(x) = p(x), which is what you would expect if the appeals court applies the same legal standard than the agency and it implements the standard showing deference to the agency.

Thus, the analysis above shows that whether the appeal process reduces the cost of error in expected terms depends on its scope (whether non-infringement decisions can be appealed) and on the possible divergence in the identification or the application of the appropriate legal standard. The implication is that the appeals process does not serve to control error costs when the following three conditions hold: (i) the appeals court must review all decisions, (ii) the appeals court is unlikely to disagree with the agency as to the nature of the appropriate legal standard, and (iii) the appeals court shows deference to the agency in the implementation of such a standard. In those circumstances, the appeal process may serve other purposes but does not increase the efficiency of the antitrust review process.

This article was originally published in EU Law Live's Competition Corner here. The views expressed are those of the authors only and do not necessarily represent the views of Compass Lexecon, its management, its subsidiaries, its affiliates, its employees, or clients.