Empirical Evidence of Hindsight Bias

Thanks to Peter Zura's Two-Seventy-One Patent Blog for uncovering "Patently Non-Obvious: Empirical Demonstration that the Hindsight Bias Renders Patent Decisions Irrational," where Gregory N. Mandel discusses an experiment in which participant mock-jurors were given a hypothetical fact scenario concerning an invention. The scenarios were each based loosely on the facts of actual issued patents that were challenged on non-obvious validity grounds in litigation and were the subject of a reported decision.

Each participant received only a single scenario in one of three different conditions. The "foresight condition" (or control condition) included all of the lead-up information and ended with the scenario character trying to solve the identified problem. The "hindsight condition" was identical to the foresight condition except that it had one additional sentence at its end which stated that the character had come up with a solution, and stated what the solution was. The "debiasing condition" was identical to the hindsight condition, but the questions following the scenario included instructions based on Model Patent Jury Instructions that informed the participant of the hindsight problem, warned him or her about it, and advised him or her not to use hindsight in answering the questions.

According to Professor Mandel,

Because the only thing that varied between the foresight and hindsight
conditions was the presence of information concerning achievement of the
invention (i.e., because all other factors were controlled for), any differences
between the foresight and hindsight groups’ responses can be attributed to the
presence of this information. Similarly, because the only thing that varied
between the hindsight and debiasing conditions was the presence of the debiasing
jury instruction, any differences between the hindsight and debiasing groups’
responses can be attributed to the presence of this instruction.

. . . As expected, participants rated inventions non-obvious significantly more frequently in foresight than in hindsight in both the baseball scenario (chi-squared=25.203, Fisher’s p < .001) and the fishing lure scenario (chi-squared=10.623, Fisher’s p < .01). Descriptive statistics are presented in Table 1. For the baseball scenario 24% (10 out of 42) of participants in the foresight condition thought that a solution to the problem was obvious, while 71% (59 out of 83) of participants in the combined hindsight conditions thought that a solution was obvious (see Table 1). Results were similar for the fishing lure scenario: 23% (9 out of 40) of participants in the foresight condition thought that a solution to the problem was obvious, while 54% (44 out of 82) of those in the combined hindsight conditions thought that a solution was obvious (see Table 1).

. . . As expected, debiasing instructions had no significant effect on judgments of obviousness in both the baseball scenario (chi-sqauared = 1.079, Fisher’s p = ns) and the fishing lure scenario (chi-squared=.785, Fisher’s p = ns). Combining the hindsight and debiasing data across both scenarios similarly demonstrated no significant effect of debiasing instructions (?2 = 1.813, Fisher’s p = ns). Participants were no more likely to consider an invention non-obvious in the debiasing condition than they were in the hindsight condition.

Among Professor Mandel's extrapolations from this, and other, data:

1. hindsight bias distorts patent decisions far more than anticipated, and to a greater extent than other legal judgments;
2. neither the Federal Circuit’s suggestion test nor the Supreme Court’s Graham requirements solve the hindsight problem;
3. the admission of secondary consideration evidence does not cure the hindsight bias;
4. jury instructions that explicitly identify and warn against the hindsight bias do not ameliorate its impact; and,
5. the hindsight problem pervades patent law to an extent not previously recognized—it biases decisions under the doctrine of equivalents, claim construction, the on-sale bar, and enablement.