Skip to main content
Category: Effects
Type: A harsher rating of harm from an act than from a matched omission
Origin: Ilana Ritov and Jonathan Baron, Journal of Behavioral Decision Making (1990); Mark Spranca, Elisa Minsk, and Jonathan Baron, Journal of Experimental Social Psychology (January 1991)
Also known as: A preference for the greater harm when it comes from not acting
Quick Answer — Omission Bias is a harsher rating of harm that comes from an act than of the same harm from not acting, when the intention and the result match. Spranca, Minsk, and Baron asked 57 students to rate a written tennis dinner. Thirty-seven rated each omission as less bad than either matched act, and one rated an omission as worse. Ritov and Baron’s vaccine question, in the same line of work, asks how many deaths from the vaccine a person will accept when the disease kills more.

​
What is Omission Bias?

Omission Bias is a harsher rating of harm that comes from an act than of the same harm from not acting, when the intention and the result are the same. It is also a willingness to accept the larger harm when that larger harm comes from not acting. A vaccine that kills fewer children than the disease, and a dinner at which a player recommends the dressing or stays quiet, are the two pictures. In the dinner, the stomach ache is the same. In the vaccine question, the death rate from doing nothing is higher.
Subjects often rated harmful omissions as less immoral, or less bad as decisions, than harmful commissions.
That sentence is the abstract of Spranca, Minsk, and Baron in 1991. Intentions, motives, and consequences were held constant. Status Quo Bias is a preference for the state already in place. On this page the omission can be the option that lets the state change. Action Bias is a preference for the move that counts as doing something, in a setting where that move is the usual one and staying still has the higher recorded success rate. Default Effect is the higher rate at which a pre-selected option is kept when doing nothing keeps it. A blank that has already been marked is a different operation from a harsher rating of an act.

​
Omission Bias in 3 Depths

  • Beginner: Take two written endings in which the guest gets the stomach ache and the host wins the match. In one, the host recommends the dressing. In the other, the guest orders it and the host says nothing. Write down what is the same.
  • Practitioner: Write the harm from acting next to the harm from not acting, and mark whether the same people are hit. In the basic vaccine item the disease kills 10 children out of 10,000. Twenty-three percent of the answers tolerated a vaccine risk of zero.
  • Advanced: The 37 of 57 is a moral rating of a story, after a rule that throws out some numerical gaps. The vaccine percentages are a policy question put to paid undergraduates. A later free-response mean has no printed median. A mail survey of magazine subscribers is another list.

​
Origin

The name is already in print in Ritov and Baron’s 1990 paper on vaccination. That paper cites the judgment experiments of Spranca, Minsk, and Baron as in press. Those experiments are the January 1991 article. Mark Spranca, Elisa Minsk, and Jonathan Baron were at the University of Pennsylvania. “Omission and commission in judgment and choice” is in the Journal of Experimental Social Psychology, volume 27, number 1 (January 1991), pages 76–105. The counts below are from the authors’ HTML version of that article. The journal PDF was not re-read cell by cell. Subjects were 57 undergraduate and graduate students, recruited by signs and paid 4.50anhour,or4.50 an hour, or 5 an hour in later sessions. They read pairs of options. One option was an omission. The other was a commission. They rated the actor’s morality, or the goodness of a decision. The authors describe the bias as an overgeneralization of a useful distinction. Omissions often come with ignorance, a weaker intention, or less effort. Those differences are missing when the actor knows both options and the intention is the same. The vaccination questions are Ritov and Baron, “Reluctance to vaccinate: Omission bias and ambiguity,” Journal of Behavioral Decision Making, volume 3 (1990), pages 263–277. The issue is October/December 1990. Both authors were at the University of Pennsylvania. The counts are from the authors’ HTML version. The journal PDF was not re-read cell by cell. The problems were modeled on the diphtheria-pertussis-tetanus vaccine. This page does not carry an injury rate for that vaccine. The article they cite for the rate was not re-read. A second result keeps the overall chance fixed and adds a risk group. Twenty-eight students took part. Four were excluded because their answers showed they had not understood the task. The means in the authors’ table are labeled N = 20. This page does not reconcile 28, 4, and 20. Twelve subjects were less willing to vaccinate when death from the vaccine was confined to a risk group than in the basic case, and 2 were more willing (p < .02). The table’s basic personal mean is 5.458 deaths per 10,000, and the vaccine risk-group mean is 4.350, against a flu rate of 10 per 10,000. A risk group for the flu moved 4 students each way (p = 1). A separate group of 22 was told that the test had to be done at birth and had not been available. Eight were less willing with the vaccine risk group, and 1 was more willing (p < .02). Six gave the maximum to both versions, and 7 gave the same lower rating to both. Ambiguity Aversion is a preference for a prize whose chance is known over the same prize whose chance is not known. The risk-group item is a relative of that preference. It is not an urn. Ilana Ritov and Jonathan Baron, “Status-quo and omission biases,” Journal of Risk and Uncertainty, volume 5, number 1 (February 1992), pages 49–61, pull the two apart. The counts are again from the authors’ HTML. The journal PDF was not re-read cell by cell. Subjects reacted more strongly to a bad outcome caused by action, whether or not the action kept the current state. They preferred inaction even when inaction was the option that changed the state. A matching task, which did not require an action, showed no consistent status-quo bias. Loss Aversion is a greater weight on a loss than on a matched gain. Ritov and Baron say that account can be kept only if the reference point is the omission, rather than the state already occupied. They also say it does not cover the cases in which someone accepts no risk at all from the vaccine.

​
Key Points

Four comparisons keep the harm fixed and move the act. A moral rating of a dinner is not a fine imposed by a court. A tolerated death rate is not a death that occurred. A feeling about a lost flight is not a preference for the ticket the person already holds. A mean rating can still favor the lower risk while a minority of comparisons go the other way.
1

Hold the harm fixed and move the act

In the tennis story, John knows Ivan Lendl is allergic to cayenne pepper, and the house dressing contains it. John can recommend the dressing, or he can say nothing after Ivan orders it, knowing a warning would make Ivan switch. The stomach ache and the match result are paired across endings. Thirty-seven of 57 students rated each omission as less bad than either matched commission. One rated an omission as worse. The authors set aside some numerical gaps when the written reason was a difference in intention. In a second story, a local witness can lie to the police or stay silent, and the person who is fined is held constant. Thirty-nine of 57 rated each omission as less bad, and none did the reverse. The scale ran from 0, not immoral, to −100. Among the students who distinguished omissions from commissions, the gap between those mean ratings ran from 1 point, for five students, to 70. The mean of those gaps was 27.6, the standard deviation 21.2, and the median 25.
2

Put the disease rate next to the vaccine rate

Ritov and Baron told 53 undergraduates that a disease kills 10 children out of 10,000. A vaccine prevents it and can kill some other children. In the basic item, 57 percent of the answers tolerated a vaccine risk from 1 to 8 per 10,000, 23 percent tolerated no risk, and 9 percent gave 9 or 10, which the authors call the maximum. Six subjects gave an answer the authors could not use, or no answer. When the children who would die from the vaccine were the children who would have died from the disease anyway, 47 percent accepted the maximum of 9 per 10,000, against 9 percent in the basic item. Sixty-eight percent tolerated a higher risk in that same-children item than in the item with the same cost, and 4 percent tolerated a lower risk (p < .001). A price of 10insteadof10 instead of 2 moved 4 of 53 toward less vaccination and 3 toward more. The rest were unchanged.
3

Let inaction be the option that changes the state

In the 1992 feeling study, 20 undergraduates ranked how bad four people felt after a written loss. The five stories were a flight, a stock holding, a course grade, a job, and a cab ride. Rank 1 was the person who felt worst. Keeping the current state by doing nothing felt less bad than changing it by acting: 16 students ranked the first as less bad in most of the five stories, and 1 ranked the other way (p < .01). The new comparison flips the act. The mean rank for inaction that lets the state change was less bad than the mean rank for action that keeps the state, in every story. Fourteen students showed that order in most stories, and 4 showed the reverse (p < .05). In a choice study with 50 students, the authors’ omission share is the percentage of conflicting pairs that favored the omission. That share was 67 percent when there was no status quo (t = 3.17, p < .005) and 71 percent when there was one (t = 4.85, p < .001). Those two shares did not differ (t = 0.67). The status-quo share did not differ significantly from half overall (t = 0.36, p = .722). The average gap, omission share minus status-quo share, was 20 percent (t = 4.33, p < .001).
4

Read a mean rating beside the reversals

Spranca and coauthors also put the student in the role of a physician, a patient, or a health official. Doing nothing carried a 20 percent chance of permanent brain damage, and a procedure carried 15 percent, or the two chances were swapped. Across 48 students who rated both options, omissions scored 16 points higher than commissions on a scale from −100 to 100. F(1, 46) = 10.2, p < .005. The twelve standard deviations run from 35.7 to 55.6, so the means are not a tight cluster. For the physician, doing nothing at a 20 percent risk was rated −38.0, and the procedure at 15 percent was rated 53.4. With the chances swapped, doing nothing at 15 percent was rated 62.9, and the procedure at 20 percent was rated −51.5. The lower chance is the better-rated option in both rows. The safer omission, 62.9, sits above the safer procedure, 53.4. In the three cases where doing nothing was the 20 percent risk, 19 of 144 comparisons favored that omission over the 15 percent procedure. In the three cases where the procedure was the 20 percent risk, 3 of 144 comparisons favored the procedure over the safer omission. Ten students produced more reversals of the first kind, and one produced more of the second (p < .02).

​
Applications

Use the two harm counts as lines you can read. Keep a moral rating inside the story that asked for one. Keep a tolerated risk inside the question that asked for a maximum.

Write both harm counts before you decide

Before you treat doing nothing as the safer choice, write the harm from the act and the harm from not acting. In the basic vaccine item the disease kills 10 out of 10,000, and the table stops at 9. Twenty-three percent of the answers tolerated none of those vaccine deaths. Nine percent accepted the maximum.

Mark whether the people are the same people

Ask whether the person harmed by the act would have been harmed anyway. When the vaccine killed only children who would have died of the disease, 47 percent accepted 9 deaths per 10,000. In the basic item, 9 percent did. In the paired comparison, 68 percent tolerated more risk on the same-children item than on the same-cost item, and 4 percent tolerated less (p < .001).

Separate a kept state from a missing act

Ask whether doing nothing keeps the current state or lets it change. Among 20 students, 14 ranked a bad outcome as less bad when inaction allowed the change than when action preserved the state. Four ranked the other way. A pre-marked form can favor the option already selected. This comparison favors the missing act.

Keep a story rating off a clinic list

The 37 of 57 is a rating of John’s morality in a written dinner, on a scale that ends at −100. It is not a fine, and it is not a share of matches thrown. The later mail survey, below, is a selected group of magazine subscribers. Its 43 of 103 is not a rate for parents in general.

​
Case Study

Fifty-three undergraduates at the University of Pennsylvania were recruited by a sign on a campus walkway and paid 5anhour.RitovandBarongavethemahypotheticalflu.Thediseasekills10childrenoutof10,000.Avaccinecosts5 an hour. Ritov and Baron gave them a hypothetical flu. The disease kills 10 children out of 10,000. A vaccine costs 2 a child and prevents the disease in everyone who receives it. Side effects kill some children, and those children are not necessarily the ones who would have died of the flu. The question was the highest side-effect risk a government should accept in a compulsory program. The printed table ran from 0 to 9 deaths per 10,000. At 0 side-effect deaths, the net drop in the death rate is 10 per 10,000, and the cost per life saved at the 2priceis2 price is 2,000. At 9, the net drop is 1 per 10,000, and the cost per life saved is 20,000.Twofurtheritemsused20,000. Two further items used 10 a child. The result in the basic item is a spread, not a single refusal. Fifty-seven percent of the answers fell between 1 and 8 deaths per 10,000. Twenty-three percent said no risk should be tolerated. Nine percent gave 9, or 10, which the authors call the maximum possible risk. Six subjects gave an answer the authors could not use, or no answer. A person whose maximum sits below 9 is refusing a vaccine risk that would still cut the number of deaths. A person at 0 is refusing every vaccine death and accepting the disease rate of 10. The 57 percent are willing to vaccinate at some positive risk. They are not a group who refused the vaccine outright. A second wording told subjects that the children susceptible to the vaccine were the same children susceptible to the flu. No child would die from the vaccine who would not have died anyway. Forty-seven percent then said the vaccine should be given at the maximum risk of 9 per 10,000. Nine percent had said that in the basic item. Sixty-eight percent tolerated more risk in the same-children wording than in the wording with the same cost, and 4 percent tolerated less (p < .001). Subjects who tolerated little or no risk in the basic item often said that vaccinating on a large scale would cause the deaths of some children. One wrote that you cannot force parents to give a child a drug or a vaccine that could cause the child to die. The authors’ reading is that these subjects count a death from the vaccine as a result of the decision, and do not count a death from failing to vaccinate in the same way, when that child would not have died anyway. The $10 price is the check on cost. Four of the 53 were less willing to vaccinate when the price was high. Three were more willing. The rest gave the same answer. The authors conclude that the tilt toward omission is not the vaccine’s cost. The fivefold change in price, and in the cost per life, did not move the typical answer. The lesson is a policy question on which many of 53 students required the vaccine to be safer than “fewer deaths than the disease,” and on which the same-children wording moved the tolerated risk up. The boundary is the room. The disease, the vaccine, and the law were hypothetical. The sample was paid undergraduates. Nobody’s child was vaccinated or left unvaccinated in the study. A later paper asks a different group about a real vaccine. David A. Asch, Jonathan Baron, John C. Hershey, Howard Kunreuther, Jacqueline Meszaros, Ilana Ritov, and Mark Spranca, “Omission Bias and Pertussis Vaccination,” Medical Decision Making, volume 14, number 2 (April–June 1994), pages 118–123, mailed 200 surveys to subscribers of a magazine that had published articles favoring pertussis vaccination and articles opposing it. The counts here are from the abstract. The PDF was not re-read. One hundred and three people replied, 52 percent. Forty-three of them, which the abstract calls 41 percent, said they did not or would not have their children vaccinated. That group was more likely to believe that vaccinating was more dangerous than not vaccinating (p < 0.001). It was also more likely to show omission bias (p = 0.004), with stated beliefs about the danger of the vaccine held constant. The abstract does not print the wording of the test items, and it does not print a death rate. Subscribers who read a magazine that argues both sides are not a sample of parents at large. Forty-three of 103 is the share of this reply pile who reported skipping the vaccine. It is not a national share, and it is not the 23 percent from the laboratory item.

​
Boundaries and Failure Modes

Omission Bias is the wrong name when a rating of a written dinner is retold as a sentence, when a free-response mean is retold as a parent’s threshold, and when a preference for the state already in place is given this page’s name. The first boundary is the form of the answer. Experiment 1 used a table that stopped at 9. A later free-response study in the same 1990 paper, with 40 subjects, asked for the smallest number of flu deaths, out of 100,000, that would lead the subject to vaccinate. When the vaccine killed 10 children out of 100,000, the mean of those answers was 2,584. When the vaccine killed 100, the mean was 4,316. When it killed 1,000, the mean was 10,693. The authors call this an immense omission bias, and they say the flu number is at least ten times the vaccine number. No median is printed. Dividing 2,584 by 10 gives the ratio of those two means. That division is not a threshold the paper prints, and it is not the typical subject’s ratio. The ratios of the three pairs of printed means fall as the vaccine number rises. A clinic, a ministry, and a manufacturer do not inherit 2,584. The second boundary is the neighboring names. In the 1992 choice study, favoring the omission did not require a status quo. When no omission was available, the status-quo share was 62 percent (t = 2.13, p < .05). When an omission was available, that share was 49 percent. Outcome Bias is a higher rating of a decision once the result is known to be good, with the odds already on the page. In the tennis story, 8 of 57 rated the behavior worse when the intended harm occurred, and none did the reverse. In the witness story the split was 6 against 1. Across both stories, 9 students showed that pattern on at least one story, and 1 showed the reverse (binomial p < .02, one-tailed). Every student who showed it on a given story also rated the commission as worse. Those cells are not Baron and Hershey’s 1988 mean gap. Action bias, on the page linked above, is the reverse feeling in a setting where the usual move is to act. Spranca and coauthors note that a few justifications in the treatment study said it was better to do something than to wait. They say that intuition might predominate when action is expected. It is not the result they report as the main one. The third boundary is the export. The jury question in the tennis story asked what award to recommend if Ivan sued for the lost 10,000andforanother10,000 and for another 10,000. Thirty-four said the penalty should be larger for a commission, 14 said the awards should be equal, none said the penalty should be larger for an omission, 7 said no penalty should be paid, and the authors say the remaining 3 did not answer clearly. Those counts sum to 58 in the authors’ HTML, against 57 students, so this page does not force them into one pie of 57. That question asks for a recommended award in a story. It is not a verdict. The causality answers are the authors’ account of why the ratings split. Of 26 students who showed the bias on the tennis story and who answered the causality questions in a way the authors could use, 22 said John played a larger causal role in the commission endings and that this justified the moral gap. Of 41 usable answers from both sides of the rating, 31 tied the moral judgment to that causal role (chi-square = 13.26, p < .001). A belief about cause is not a count of stomach aches. The pertussis mail survey holds beliefs about danger constant and still finds a difference at p = 0.004. It does not say what share of that difference would remain in a different magazine, or in a city.

​
Common Misconceptions

Three retellings travel with the quiet ending. Doing nothing is treated as the bias in every setting. The missing act is treated as the same thing as keeping the current state. The laboratory spread is treated as if almost everyone refused any vaccine risk.
In the treatment ratings, the lower chance was still rated better on average. The 16-point gap is omissions over commissions across both chances, not a mean preference for the 20 percent risk. The reversals were 19 of 144 comparisons in one direction and 3 of 144 in the other. In the vaccine table, 57 percent of the answers accepted a risk from 1 to 8 per 10,000. The group at zero is 23 percent, not the whole sample of 53.
Status quo bias is a preference for the state already occupied. The default effect is a higher rate of keeping the option a form has already marked. In the 1992 feeling ranks, action that kept the state felt worse than inaction that let the state change: 14 students against 4. In the choice study, 67 percent of the conflicting pairs favored the omission when neither option was the status quo. A pre-checked box can favor doing nothing because nothing keeps the mark. The pattern on this page can favor doing nothing when nothing changes the state.
The 23 percent, the 57 percent, and the 9 percent are answers from 53 undergraduates about a hypothetical flu and a hypothetical law. The same-children wording moved the share at the maximum from 9 percent to 47 percent. The 1994 mail survey is 103 replies from subscribers to one magazine, and 43 of those replies reported skipping pertussis vaccine. The test items are not printed in the abstract this page used. Neither figure is a share of parents in a country.
Omission Bias sits beside a preference for the state already in place, beside a preference for the usual move, beside a mark that doing nothing keeps, beside a known chance set against an unknown one, and beside a rating that follows the result.

Status Quo Bias

A preference for the option already in place. In the 1992 choice study, omissions were favored when inaction changed the state. The status-quo share did not differ significantly from half overall.

Action Bias

A preference for the move that counts as doing something, when staying still has the higher recorded success rate. Spranca and coauthors note the opposite intuition when action is the expected move.

Default Effect

A higher rate of keeping the option that doing nothing selects. A pre-checked box can favor inaction. The vaccine item has no pre-checked box. The question is a maximum risk.

Ambiguity Aversion

A preference for a known chance over an unknown chance with the same prize. The 1990 risk-group item keeps the overall chance fixed and still lowers willingness to vaccinate.

Outcome Bias

A higher rating of the same decision when the result is good. In the tennis story, 8 of 57 rated the behavior worse when the harm occurred. That cell is not the 1988 mean of 0.70.

​
One-Line Takeaway

Before you treat a harmful act as worse than doing nothing, write the harm from the act, the harm from not acting, and whether they fall on the same people.