Skip to main content
Category: Effects
Type: A stated willingness to pay that does not rise with the number of birds the questionnaire says the program would save. The score is a hypothetical yearly household amount, in higher prices, not cash handed over. On the cleaned bird tables the means are $80, $78, and $88. The median is $25 in every version
Origin: William H. Desvousges, F. Reed Johnson, Richard W. Dunford, Kevin J. Boyle, Sara P. Hudson, and K. Nicole Wilson, Measuring Nonuse Damages Using Contingent Valuation: An Experimental Evaluation of Accuracy. The PDF read here is the RTI Press second edition, publication BK-0001-1009, DOI 10.3768/rtipress.2009.bk.0001.1009. It was first published in 1992 as Research Triangle Institute Monograph 92-1. The second edition says the original text is unchanged. Boyle’s foreword and the preface to the second edition are new. Exxon funded the experiments after the Exxon Valdez spill of 24 March 1989. The preface says the results are the authors’ opinions, not necessarily Exxon’s. Kahneman (1986), the NOAA panel report, and later reviews cited in the foreword were not opened
Also known as: Scope insensitivity, the foreword’s phrase for whether a value responds to the size of the injury. The 1992 chapters call the test theoretical validity. They do not print “scope neglect.” This page is not Duration Neglect and not the Identifiable Victim Effect
Quick Answer — Scope Neglect, as this page uses the name, is a stated yearly bid that barely moves when the number of birds saved grows by powers of ten. In the 1992 mall study, separate groups were asked what their household would pay each year, in higher prices, to keep about 2,000, 20,000, or 200,000 migratory birds from drowning in waste-oil ponds. On the cleaned tables the means are $80, $78, and $88. Every median is $25. The authors’ preferred tests do not find a higher bid for the larger number. They also print two weaker exceptions, and they say the study does not show why the bid stayed flat.

​
What is Scope Neglect?

Scope Neglect is a stated willingness to pay that does not rise when the questionnaire’s count of birds saved rises from 2,000 to 20,000 to 200,000.
Our results for the migratory-waterfowl tests show that stated WTP values are statistically constant regardless of whether 2,000, 20,000, or 200,000 bird deaths are prevented.
Desvousges and colleagues print that sentence after the bird tests in Chapter 5. WTP means willingness to pay: the most the household says it would accept in higher prices each year. The three numbers are three versions of one open-ended question. Each person saw one version. The cleaned means are $80, $78, and $88. The sentence calls those stated values statistically constant. It does not say every test in the chapter was a null, and it does not say anyone paid those dollars. A useful picture is three copies of the same form, with one blank filled differently. The program, the wire-net covers, the payment in higher prices, and the map of the Central Flyway stay put. Only the example death count changes, and with it a short share of 8.5 million birds. Duration Neglect is a different page. There, the length of an episode gets little weight in a later rating. Here, the count of birds gets little weight in a stated yearly price.

​
Scope Neglect in 3 Depths

  • Beginner: Three groups saw 2,000, 20,000, or 200,000 birds. The cleaned means were $80, $78, and $88. The median was $25 on every form.
  • Practitioner: Before you average a survey, write whether each person saw one count or several, and whether the figure you quote is the mean or the median. In this study those two summaries do different work.
  • Advanced: The linear model means are $59, $59, and $71, and those gaps are not significant. Two log comparisons reach the 10 percent line. A likelihood-ratio test says the coefficient vectors differ. A flat mean is not a flat model, and a mall sample is not a national damage figure.

​
Origin

Scope Neglect, on this page, is the name for that flat stated bid. The 1992 monograph does not use the name. The experiments were funded by Exxon as research tied to the 1989 Exxon Valdez oil spill. The first-edition preface says Exxon wanted to test the accuracy of contingent valuation for nonuse values. Contingent valuation is a survey that asks people what they would pay for a change that is not sold in a store. Nonuse, in this book, is a value held by people who do not use the resource. The preface says the opinions are the authors’. It also says Bill Desvousges presented some of the results at a Washington symposium on 2–3 April 1992. The monograph is the longer account. Boyle’s 2010 foreword, which is new to the second edition, calls the bird design and a companion oil-spill design tests of scope. He writes that at the time of the spill, Kahneman (1986) had argued, with what the foreword calls weak empirical evidence, that contingent-valuation estimates might not be sensitive to the extent of injury. That 1986 paper was not opened. The foreword’s hypothesis is the economic one: a larger injury should draw a larger willingness to pay. The 1992 text names the same idea theoretical validity. Hypothesis 1 says WTP estimates increase for higher levels of natural resource services. The bird question is open-ended. People wrote a dollar amount. The oil-spill experiment, a different commodity in the same book, used both an open-ended question and a yes-or-no question. The foreword says the NOAA Blue Ribbon Panel strongly recommended referendum, or dichotomous-choice, questions, and that this book’s experiments use the between-subject approach. The panel report was not opened. Base Rate Neglect is a different mistake, about setting aside a stated rate in a probability problem. These forms are not that problem.

​
Key Points

Four splits keep a cleaned mean, a median, a model, and a design from becoming one slogan.
1

The score is a stated yearly price, and the cleaned mean is not the median

The question asks for the most the household would agree to pay each year in higher prices for wire-net covers. The form says oil companies would pass the cost on, and that higher petroleum prices would raise the prices of most other things the household buys. No one handed over cash, and no incentive was paid for filling in the form. Table 5-1, labeled censored data, prints means of $80, $78, and $88 for 2,000, 20,000, and 200,000 birds. The standard deviations are $187, $132, and $166. The median is $25 in all three columns. The modes are $0, $100, and $100. The ranges are 0–1,550, 0–1,000, and 0–1,000. The samples in that table are 288, 286, and 281.
2

Censored, here, means the responses left in after cleaning

Completed bird questionnaires were 398, 408, and 399, totaling 1,205. Table 3-6 says about 70 percent of the open-ended bird responses were usable. Included zeros are 146. Included amounts above zero are 709. Those two rows sum to 855, which is 288 + 286 + 281. Dropped rows, for the bird forms together, are 95 protest zeros, 18 amounts over the income screen, 31 outliers, and 206 answers that were not a number. Protest reasons include “my household should not have to pay,” not enough information, higher prices as a payment vehicle, doubt that the covers would work, and being unable to name a dollar amount. The income screen drops amounts greater than 25 percent of income. The body text also says amounts of $10,000 or more were eliminated. The table note mentions a $12,000 amount. This page does not merge those two ceilings. Outliers were flagged with the Belsley, Kuh, and Welsch method. All but two of the 31 bird outliers had a stated amount of $1,000 or more.
3

The preferred tests do not find a larger bid, and two other tests do not agree

Shapiro-Wilk statistics of 0.43, 0.60, and 0.54 say the amounts are not normal, so the authors use rank tests. Wilcoxon p-values are 0.178, 0.457, and 0.145. Permutation p-values are 0.425, 0.185, and 0.288. None rejects equal distributions. Tobit means from the linear final model are $59, $59, and $71, with standard errors $11, $7, and $10, and sample sizes 267, 271, and 264. One-tailed z statistics are −0.02, −0.98, and −0.82. None is starred. Log-model means are $81, $80, and $123, with standard errors $16, $16, and $29. The z statistics are 0.06, −1.30, and −1.26. The last two are marked significant at the 10 percent level for a one-tailed test. The authors prefer the linear form and say they fail to reject no difference in the means. A likelihood-ratio test on the coefficient vectors is a different result: chi-square 50.6 for the linear models and 152.9 for the log models, both significant at the 1 percent level. Footnote 8 says the effective sample in Table 5-7 is about 400. The table prints 267, 271, and 264. This page does not repair that footnote.
4

Each person saw one count, and the form also printed a share

The 2,000-bird page says that in 1989 about 2,000 birds died in the ponds, “much less than 1 percent” of 8.5 million in the Central Flyway. The 20,000-bird page says “less than 1 percent.” The 200,000-bird page says “about 2 percent.” The count was not hidden. Two of the three shares still sit under 1 percent while the counts differ by a factor of ten. Boyle’s foreword says this is a between-subject, external test of scope: each person sees one level of injury and is not shown the other levels side by side. The foreword says that design asks for a comparison people are not asked to make when they shop. It does not rerun the survey.

​
Applications

These four moves follow splits the monograph reports. They are not a claim that a fund, a class, a household, or another year copies $80, $78, or $88.

Write the count and the share apart

The form prints both a death count and a share of 8.5 million. The 2,000 and 20,000 versions both use a “less than 1 percent” line, one of them with the word “much.” A memo that says only “under 1 percent” has dropped the blank that actually changed. Ask which number was on the page.

Quote the median next to the mean

The cleaned means are $80, $78, and $88. The median is $25 in every column, and the standard deviations are larger than the means. A single average can sound like a typical household. On this table the typical stated amount is the median, and it does not move at all.

Say whether anyone saw two scopes

Each mall respondent received one version. The foreword calls that a between-subject test. A later form that places 2,000 and 200,000 on the same page is a different design. This book does not report that comparison for these birds.

Keep the mall mean off a national bill

The authors say the samples were built to compare versions, not to produce a damage estimate for a named population. Atlanta shoppers in May and June 1991, with no cash payment, are the people in the table. A restoration invoice is not Table 5-1.

​
Case Study

The counted case is the migratory-waterfowl experiment. Research Triangle Institute fielded it with Jackson Associates in two suburban Atlanta malls. Southlake Mall, south of Atlanta, produced 1,923 completed questionnaires of every version, bird and oil spill together, from 17 May to 5 June 1991. Lakeshore Mall, in Gainesville, produced 889 from 22 May to 5 June 1991. Recruiters kept people over 19 who were not full-time students. Nobody was paid. A questionnaire took about 10 to 12 minutes. Bird forms were randomly assigned: 398, 408, and 399 completed. The authors say the malls were chosen because Atlanta is outside the Central Flyway and not on a coast, so a nonuse value would dominate. They also say the design is a comparison across forms, not a number to generalize. The form states that remote parts of Texas, Oklahoma, and New Mexico hold over 250,000 waste-oil ponds, 10 to 100 feet across, and that birds land, sink, and drown. It names mallards, pintails, white-fronted geese, snow geese, and greater sandhill cranes. It says the government is considering wire-net covers, identification of ponds, and monitoring by the U.S. Fish and Wildlife Service, paid for by taxes on oil and gas producers that would show up as higher prices. The only planned change across the three bird forms is the example count, and the share of 8.5 million that goes with it. Drawings of the nets were dropped after pretests confused people. The income question asks about 1990. Table 5-1 then shows the cleaned bids. Means $80, $78, and $88. Medians $25, $25, and $25. The rank tests do not reject equal distributions. The linear tobit means, $59, $59, and $71, are not significantly different. The log means, $81, $80, and $123, differ at the 10 percent line when 200,000 is compared with either smaller count. The likelihood-ratio test says the equations are not the same shape. The authors’ summary is that stated willingness to pay is statistically constant across the three counts, and that this fails the theoretical-validity hypothesis that a larger saving should be worth more.
A fundamental challenge in many contingent-valuation tests of scope is that between-subject, external tests of scope are conducted. This is the approach in the experiments reported in this monograph.
The lesson inside this file is narrow. On these forms, a hundredfold rise in the stated death count did not produce a reliably higher cleaned mean, and it did not move the median. The same chapter says the study does not determine the cause. A picture of one oiled bird is not a measured result here. Affect Heuristic is a different page, and this questionnaire did not score a feeling.

​
Boundaries and Failure Modes

Scope Neglect, as this PDF lets us count it, is a set of stated yearly prices from adults in two Atlanta-area malls in late spring 1991. Two uses overshoot the book. One exports $80, $78, $88, or $25 onto a fund, a class, a household, or another year. The other treats the authors’ summary sentence as if the log model, the likelihood-ratio test, and the footnote had not been printed. The linear gaps are small relative to these standard errors. Footnote 8 says that if the means and variances stayed put and the samples grew, the linear contrast of 200,000 birds with the other two versions would become significant if the 2,000-bird sample rose to about 1,000 and the 20,000-bird sample rose to about 700. Those larger samples were not collected. The best pseudo-R² the chapter mentions is 0.14, for one full linear model. A flat mean can sit on a noisy equation. The other scope test in the same book is not flat in the same way. Open-ended questions about reducing the effects of small oil spills, versus small and large spills together, give cleaned means of $144 and $95. The standard deviations are $239 and $157. Both medians are $50. Both modes are $100. The samples are 275 and 282. The executive summary says the small-spills estimates exceed the all-spills estimates. A larger program drew the smaller mean. That reversal is a different commodity and a different pair of forms. It is not a second bird number, and the yes-or-no oil-spill results are not copied here. The foreword, written in 2010, says scope tests are still hard, and that choice experiments, where the same person sees more than one level, have sometimes shown more sensitivity. Foster and Mourato (2004) and the other papers named there were not opened. No coefficient from them is printed here. The bird instrument remains the open-ended mall form. Zero-Risk Bias is a preference for driving one small risk to zero. No risk was priced that way. Availability Heuristic is about what comes to mind easily. This monograph does not score that, and it says it does not identify the cause of the flat bid. A last limit is the neighbor. Identifiable Victim Effect is a larger gift to one named person than to a page of statistics, counted in dollars from a $5 payment. These bird forms name no child, show no photograph, and ask for a yearly price. The two pages do not share a sample.

​
Common Misconceptions

Three readings turn a cleaned mean, a preferred model, or a printed count into a different study.
Table 5-1 is a cleaned distribution of hypothetical yearly amounts, written as higher prices. The means are $80, $78, and $88. The median is $25. Nobody in the method section was paid to participate, and the donations were not collected. “Average contribution” is a looser later phrase. It is not what this table records. The 2,000-bird mode is $0. The other two modes are $100. Those modes are not payments either.
The rank tests do not reject equal distributions. The linear z statistics, −0.02, −0.98, and −0.82, are not significant. The log comparisons of 200,000 birds with 20,000 and with 2,000 are marked at the 10 percent level, z = −1.30 and −1.26. The likelihood-ratio statistics, 50.6 and 152.9, reject equal coefficient vectors at the 1 percent level. The authors still conclude that the means are not reliably different, because they prefer the linear specification. That conclusion is printed. So are the exceptions. Quoting only the flat sentence drops the log stars and the coefficient test.
Each version prints its count in the story and again in the question: about 2,000, about 20,000, or about 200,000 deaths each year. Each version also prints a share of 8.5 million. The first two shares are both under 1 percent. The third is about 2 percent. The number was on the page. What the person did not see was the other two versions. Between-subject is not the same claim as “the count was omitted.”
These six pages sit beside Scope Neglect. Each one asks a different question.

Identifiable Victim Effect

A named person draws a larger gift than a page of statistics. The score there is dollars from $5. The score here is a stated yearly price for covers.

Duration Neglect

The length of an episode gets little weight in a later rating. These forms vary a death count, not the length of a film or a pain.

Base Rate Neglect

A stated rate is set aside in a probability problem. The flyway share on these forms is not that problem.

Affect Heuristic

A feeling is used as information about a value. This monograph does not score a feeling, and it does not name the cause of the flat bid.

Zero-Risk Bias

A preference for wiping out one small risk. No risk was driven to zero. The question asks for a yearly price.

Availability Heuristic

What comes to mind easily is judged more common. A single image can be easy to picture. This study did not score availability.

​
One-Line Takeaway

Before you quote one average from a scope question, write the count each person saw and put the median beside the mean. In this study the medians were all $25, and the cleaned means were $80, $78, and $88.