Moonshine &$J» Why the Remains an Icon of Evolution

The peppered moth evolved a dark form in response to soot covering its habitat during the industrial revolution. Or did it? There has been speculation that the data were faked or manipulated. An investigation shows that this criticism is without foundation.

MATT YOUNG and IAN MUSGRAVE

ernard Kettlewell did not cheat—or, more precisely, the evidence does not support the insinuation, widely Brepeated on the Internet, that he cheated. Kettlewell was a distinguished naturalist whose studies on predation in peppered moths were a landmark in demon- strating natural selection in the wild. His studies are widely quoted and often used in textbook accounts of natural selec- tion. The journalist Judith Hooper (2002) strongly suggests, however, that Kettlewell fraudulently altered the results of his famous studies, and others have uncritically accepted her suggestion. Reviews of Hooper's book in the scientific liter- ature are at best mixed (Coyne 2002, Grant 2002, Shapiro 2002), and experts on moth behavior remain convinced that

SKEPTICAL INQUIRER March/Apr,I 200S 23 know that birds had ultraviolet vi- sion, which his observers lacked, but nevertheless got a good •X, K correlation between camou- flage and predation. Later ^^a^S i ' Y V" research has shown that the moths are ^^•kj camouflaged in the ultraviolet as well as in the visible (Majerus et al. 2000, Musgrave 2005). 4. Geographical distribution. He noted that the distribution of the melanic moths in the country closely matched the areas of industrialization (Bishop and Cook 1975).

Figure 1. The pale or light-colored form (top) and the dark or melanic We will discuss only the release-recapture experiments form (bottom) of the peppered moth Biston betularia. The wings span 4-5 reported in Kettlewell's 1955 article, because these are the cm. Photo by Bruce S. Grant, Journal of Heredity, vol. 93, March-April 2002 (cover). Copyright ©2002 by the American Genetic Association, experiments that are under fire and because (unlike Kettlewell's reprinted with permission. critics) we can bring quantitative tools to bear on the analysis. For a more general analysis, see 's "Fine Tuning the the story of the peppered moth is sound (Cook 2000, 2003; Peppered Moth" in Evolution 53 (1999): 980-984 and Ian Grant 1999; Majerus 1998, 2002; Mallet 2004). More perti- Musgraves "Paint It Black: The Peppered Moth Story" in Icons nently, Kettlewell's data are completely consistent with normal of Evolution, which will be published by the Education experimental variation, and Hooper's insinuations are groundless. Committee of the Society for the Study of Evolution in 2005. Kettlewell's Experiments Day Date- Number released Number recaptured What did Kettlewell do, and why does Hooper think he fudged 0 June 25 32 - his data? Beginning in the mid-1800s, successive generations of 1 June 26 0 5 peppered moths (Biston betularia) in Britain gradually dark- 2 June 27 59 2 ened in response to the air pollution in the industrialized parts 3 June 28 63 4 of the country (figure 1). Specifically, a genetically determined 4 June 29 0 9 dark, or melanic, form of the moth replaced the lighter form as 5 June 30 102 2 industrial pollution killed lichens on the bark of trees and also 6 July 1 114 23 coated the bark with a layer of soot. The effect has come to be 7 July 2 98 34 known as , and its existence is not in dispute 8 July 3 83 29 (see Forrest and Gross 2004, 107-111 for a review). 9 July 4 79 25 Kettlewell (1955, 1956, 1959) showed that the melanic 10 July 5 0 16 form of die moth predominated primarily because of predation Totals 630 149 by birds. He did not think diat predation was the only cause of industrial melanism and in fact speculated as to the relative *Recapture s reported as June 26 took place between the evening of June 25 and the morning of June 26, and so on. strengths of other causes. Briefly, he performed a number of experiments (Kettlewell 1959, Grant 1999, Musgrave 2005): Table 1. Kettlewell's 1955 release-and-recapture data.

1. Release-recapture experiments. Kettlewell marked and released both light-colored and melanic moths early in the morning, and 40 -I ys=0.25x recaptured some the next night in both -o • pheromonc and light traps (using mercury- r2=0.80 vapor lamps). In polluted woods, he and his | 30- • ^^. assistants recaptured more melanic moths a. • than light-colored (1955, 1956), whereas in ui ^-^* unpolluted woods they recaptured more light- | 20- h i ^^•"""« colored than melanic (1956). • f 10- ^^-^"""^ • 2. Direct observation (1955, 1956) and film- 3 ^^^m ing (1956). Kettlewell and others observed Z ^^-^ • birds eating moths directly off trunks of trees. o ! ^^»— () 50 100 3. Camouflage. Kettlewell v i s u a l l y ranked the effectiveness of camouflage of moths on dif- Number Released ferent backgrounds and compared the effec- tiveness of camouflage with predation rates Figure 2. The number of moths Kettlewell recaptured as a function of the number he released both in an aviary and in the field. He did not the previous night. The relationship is linear with high correlation.

2 4 Mjrch/Apnl 2005 SKEPTICAL INQUIRER Kettlewell reported releasing and recapturing moths during increase sharply after one day of inactivity. Biological field an eleven-day period in 1953. His data are reproduced in table data, however, display significant random variation, and the 1 (1955, 332). The numbering of the days is ours. eye often infers patterns in random data, so we performed a Hooper has noted that the number of recaptures increased quantitative analysis to see whether Kettlewell's data are what sharply on July 1, the same day that Kettlewell's mentor, E. B. we would expect given normal experimental variations. Ford, sent him an encouraging letter. Ford's letter commiserated with Kettlewell for the 40 -i low recapture rates but suggested that the ) data would be worthwhile anyway. The let-

ter is unremarkable, and two facts militate o against a finding of fraud. First, Kettlewell finished collecting data in die wee hours of (Percent 2 20- the morning and dierefore could not have ro F

received the letter before collecting his data e

on July 1. Second, he markedly increased the o number of moths he released on June 30, Captur V the day before the letter was mailed, not July 1 1 1. Additionally, as Hooper admits, he con- () 2 4 6 8 10 tinued to release a high number of moths Time (Days) after June 30. Not surprisingly, he also recaptured more moths: more moths Figure 3. Kettlewell s recapture rate as a function of time, ignoring the two days following zero releases. Note that, owing to the vagaries of graphing programs, the days are not numbered released, more recaptured. correctly. Indeed, figure 2 plots recapture rate as a function of the number of moths released on any day. The line is Specifically, we constructed a mathematical model to see how a line of best fit constrained to pass through the origin on the well it describes Kettlewell's data. assumption that no moths are recaptured if none are released. Figure 2 shows that the recapture rate is very nearly a linear func- Mathematical Model 1 tion of the number released. The square, r, of the correlation Kettlewell recaptured most of his moths after they had been in coefficient is 0.80 and suggests that most of the variation of the the wild for only one day, but he recaptured some after two number of recaptures is accounted for by variation of die num- days. Let us therefore define a one-day recapture rate, R,, and ber of releases. T h e fit improves only slighdy if the line is not a two-day recapture rate, /?>, as the fractions of moths recap- constrained. tured after one and two days in the wild. Kettlewell reported Why did Kettlewell release more moths beginning no three-day recaptures. on June 30? He released both moths he had reared We may estimate the two-day recapture rate and moths he had captured. Because the moths by looking at the four moths captured were just hatching, he had limited control of on days 2 and 5. No moths were the number he could release on any given released on the preceding days, day. There is no reason to suspect that but two days before, a total of the increased numbers of releases 63+32=95 moths had been reflect anything other than the num- released, so /?2=4/95, or approxi- ber of moths that were available. At mately 4 percent. The overall recapture any rate, Ford's letter could not have rate is given in table 1 in the row labeled "Totals" influenced his decision to release more and is R-149/630, or approximately 24 percent. The moths because it arrived after Kettlewell's first big release on one-day recapture rate is the difference June 30. (R\=R— R 2) between the two val- Still, his recapture rate, as well as the absolute number ues, or about 19 percent (the of moths recaptured, increased from 12 percent over numbers do not add the first three days of his experiment to 26 percent exactly because of over the last three days. More pointedly, if we round-off error). R2 is plot his recapture rate as a function of time, as very neariy equal to the in figure 3, we find what looks to the eye as a square of /?i, as we would sudden increase. WiM '••* expect if the model is appropriate. Figure 3 omits those days, June 27 and Our mathematical model is straightfor- June 30, that were preceded by zero releases. It is ward: The number. A'.,, of moths recaptured on the rth night hard to make much out of a mere eight data points, but die is equal to die number, N.,, of moths released the day before recapture rate certainly appears to the casual observer to rimes die one-day recapture rate, /?„ plus a similar term, the

SKEPTICAL INQUIRER March/April 2005 25 through the hole, and repeat the experiment many times. • t J** **• Suppose that the average number of marbles that fall through the hole is M. You will not count M marbles every time you perform the experiment; to the contrary, the number will vary about M and very possibly will never exactly equal M. Thus,

number, A^2, of moths we talk of the probability p that any one marble passes through released two days before times the the hole and set it equal to the ratio MlN. The mean number 2-day recapture rate, /?,. Specifically, of marbles that pass through the hole is equal to Np. How much will any one toss differ from Mi The number of Nr„=RiN^R2Nr2 (i) marbles that pass dirough die hole is described by a binomial dis- The results of a calculation based on this model are shown tribution. The standard deviation of M is therefore 0=^lNp(\—p). as the solid curve in figure 4. Note that we have made no arti- On approximately nineteen tosses in twenty, you will record a ficial assumptions, such as adjusting the recapture rates to get number diat is between M-2o and M+2o, so 2(T is most com- a good fit to the data, in constructing figure 4. monly used as a measure of uncertainty. The points in figure 4 are Kettlewell's data, and the solid The uncertainty 2

H 40 - — should not be especially surprised by any 3 s. 3 number unless it is much less than twelve or 1 ° - much more than diirty-six. Thus, the day- cc mS^' ' "*"1i^ * ~~-T to-day variation in an experiment such as •' <) 5 10 moths and recaptured M modis. Because die Time (Days) two-day recapture rate is small, die statistics are essentially the same as diose of die mar- Figure 4. A mathematical model fitted to Kettlewell's data. The symbols are the data points, the ble example. The number M is highly vari- solid curve is the model, and the error bars are the 95 percent confidence intervals. The dashed curve is the model corrected for exposure to moonlight. able, as we have seen; on anodier day, he might capture a substantially different num- uncertainty of the data points. The standard uncertainty ber, even if conditions were unchanged. The standard uncer- is a number that tells us, in this case, how much variation we tainty u estimates die probable variation of M quantitatively might expect if we repeated the experiment many times. (ISO 1993). It is given by «=VNR(\-R), where die recapture By way of introduction, suppose that you toss N marbles at rate, R, replaces die probability, p. We apply diis formula to each a hole in a table. Count the number of marbles that fall data point, using die overall recapture rate, R for most days but

die two-day recapture rate, R2, for the two days that were pre- Matt Young is Senior Lecturer in Physics at the Colorado School of ceded by zero releases. Mines and a former physicist with the U.S. National Institute of The result is shown in figure 4 as a series of error bars. The Standards and Technology He is the author of No Sense of error bars represent ±2u, an interval called the 95 percent Obligation: Science and Religion in an Impersonal Universe confidence interval. If we take a single measurement, dien (1st Books Library, 2001) and co-editor of Why Intelligent we may estimate that the true value (the average of a Design Fails: A Scientific Critique of the New great many measurements) falls within the error bars, Creationism (Rutgers University Press, 2004). widi 95 percent probability. Inasmuch as the E-mail: [email protected]. Ian Musgrave, model (the solid curve) passes through the author of a chapter in Why Intelligent virtually every error bar, it may be said Design Fails, is a lecturer and researcher to be a nearly perfect fit to the data, in the Department of Clinical and however poor it might appear in Experimental Pharmacology at the » die absence of error bars. In University of Adelaide, one of the few addition, we will use another places on Earth devoid of peppered moths. statistical technique to show below E-maib [email protected]. that die models fit the data well.

2 6 March/April 2005 SKEPTICAL INQUIRER The points on days 7, 8, and 9 lie noticeably above die lation that suggests a slight curve. If die data were completely unbiased, dien we increase of capture rate would expect about a 50-50 chance diat any one of diose around die full moon. In points lay above die curve. The odds that diree consecutive addition, when they checked points lie above die curve are 1 in 8—exactly the same as the new moon against the full die odds against tossing three heads in a row and by no moon, they calculated a small, means improbable enough to base a charge of cheat- barely significant increase, which ing. Even if five points lay above die curve, die odds they discounted. Possibly the effect is against would be 1 in 32, again, not very impressive due to the presence of streetlights, to which in its improbability. Additionally, two consecutive they refer obliquely, and which may attract data points lie noticeably below die curve. moths away from the stronger mercury vapor light of In summary, die last five of Kettlewell's data the trap only when the moon is dark. At any rate, they con- points are higher man die first five. This meager fact, clude that moonlight does not affect capture rates. Kettlewell combined widi die anecdotal evidence of Ford's letter, is all that led worked on clear days only; we do not think that die conclusion Hooper (2002) to infer that Kettlewell cheated. In reality, the tim- of Clarke and colleagues is necessarily pertinent. ing of Ford's letter belies Hooper's inference, and Kettlewell's data Thus, we examined Kettlewell's data in hope of quantifying are completely consistent with normal experimental variation. the effect of the moon on his recapture rates (1955, 332, table 5). We obtained data that gave the moon's magnitude (an Moonlight astronomical term related to its brightness) and the dutation The differences between the data and the curve are not significant; the observed varia- 4 tion very probably is the result of chance. It ? 1 y=-2.21x+3.02 is, however, possible that the some of the a 2 deviation from the curve is "real"—that is, £ 3 - .••«• r=0.75 due to some systematic effect, or systematic error, not due solely to random error. It is 5 2 - • very hard, unfortunately, to track down a e • ~~~~~~------t^^^ source of systematic error when that error is • itself less than the standard uncertainty of 1 the data set; the systematic error is said to be Recaptur l lost in the noise. (). 0 0.5 1.0 Hooper tells us that the weather was stable Relative Exposure and could not have accounted for the increase in the recapture rate (though her description Figure 5. Kettlewell's daily recapture rate as a function of relative exposure to moonlight. The suggests somewhat variable winds). We have, equation is a line of best fit that relates exposure to recapture rate. nevertheless, a strong candidate that can account for the system- during which the moon was visible each night during atic deviations of our simple model from the curve: the phase of Kettlewell's experiment. We plotted Kettlewell's daily recap- the moon. Shapiro (2002), in his review of Hooper's book, sug- ture rate as a function of die exposure to the moon (the prod- gests that moonlight interferes with moth trapping, a possibility uct of brightness and time). We made no effort to control for that Hooper and her informant, biologist Ted Sargent, should the elevation of the moon. The result is shown in figure 5, have investigated. The moon was full on June 27 (diat is, die which plots Kettlewell's daily recapture rate as a function of night of June 26—27). By July 2, die moon was five days past full lunar exposure normalized to the value 1 on the night of the but visible for only pan of the night. Thus, the total exposure to full moon. The equation in figure 5 is the equation of the line the moon—die product of illuminance (brightness) and time— of best fit to the data. The daily recapture rate rises by a factor was approximately one-quarter what it was during the full moon, of 3 as the brightness of the moon decreases. (We could and it dropped steadily over the next few days. perform a similar calculation using Kettlewell's total captures Clarke and his colleagues (1990) have investigated the effect [1955: 333, table 6], but such a calculation is complicated by of the phase of the moon on capture rates of peppered moths in the fact diat die modis emerge from their cocoons haphaz- light traps in a single environment over thirty years and con- ardly, whereas the recapture rate is based on a known distribu- cluded that the moon does not affect capture rates. tion of released moths. Still, the calculation based on total cap- Unfortunately, theirs was a retrospective study, and diey did not tures yields much the same result as that outlined below.) record weather data—diat is, did not control for cloudy or rainy Using the line in figure 5, we adjusted die calculated recap- days. They averaged the data over five-day periods surrounding ture rates according to the equation: the full moon and did not use die actual exposure to the moon- light (as defined above). All of these factors will reduce the cor- 0.3017-0.2215 E relation between capture rates and exposure to moonlight. Even g=R (2) so, they calculated a small but not statistically significant corre- R

SKEPTICAL INQUIRER March/Apm 200S 27 where R represents the forms (Grant 1999), as well as independent replications of nightly recapture rate Kettlewell's landmark experiments (Grant 1999, Majerus used in die model that led 1998). In other words, an enormous body of evidence to figure 4, ^?" is the nightly supports Kettlewell's conclusion. Even if Kettlewell's recapture rate modified to include release-recapture experiments were ruled out, we the effect of lunar exposure, R is the would still be forced to conclude that industrial average daily recapture rate, and E is the melanism is die result of natural selection due to nightly exposure to moonlight. The result of the bird predadon, possibly among other causes. calculation is shown in figure 4 as the light, dashed curve. Thus, diere is no foundation for assuming that Kettlewell's data were manipulated. The variations in his data are no more dian the uncertainties associated witJi sampling Hooper's claims are moonshine; and other factors, possibly including expo- they are based on a lack of understanding of sure to die moon. It is an irresponsible leap to accuse a distinguished naturalist of fraud Kettlewell's experiments in particular on die basis of a single letter and a wholly and experimental science in general. imperfect, offhand analysis of his data. The peppered moth properly remains a valid paradigm—no, an icon—of evolution.

Copyright ©2005 by Matt Young and Ian Musgrave. All rights reserved. It demonstrates a somewhat better fit to the data than the solid curve, especially during the first few days. Acknowledgements Finally, we used a goodness-of-fit test (a x' [chi-squared] test Matt Young is indebted to Pete Dunkelberg and Bruce Grant for with nine degrees of freedom [Feinstein 2002]) to compare the helping him understand the uncertainties of field work in biology. model's predictions with the observed data. Neither model was Laurence Cook and Nicholas Matzke reviewed the paper and made significantly different from the data; that is, the model fits the many helpful suggestions regarding both clarity and content. data very well. Specifically, the statistical significance, or P-value, References was P=0.\7 for die initial model and P=0.75 for the model cor- rected for moonlight. If the model and the data had been differ- Bishop, J.A., and Laurence M. Cook. 1975. Moths, melanism, and clean air. Scientific American 232:90-99. ent, then Pwould have been 0.05 or less (that is, P<0.05 implies Clarke, Cyril A., Frieda M.M. Clarke, H.C. Dawkins, and Susannah Kahtan. a statistically significant difference). The /'-values therefore con- 1990. The role of moonlight in the size of catches of Biston betularia in firm what we already know from the uncertainty analysis. West Kirby, Wirral, 1959-1988. Bulletin of the Amateur Entomologists' Society 368: 19-29. Instead of asking why Kettlewell's recapture rate was high Cook, Laurence M. 2000. Changing views on melanic moths. Biological on July 1 and thereafter. Hooper should have asked why it was Journal of the Linnean Society 69: 431 - 4 4 1 . so low on June 30 and before. . 2003. The rise and fall of the Carbonaria form of the peppered moth. The Quarterly Review of Biology 78(4): 1-19. Coyne, Jerry. 2002. Evolution under pressure. Nature AH: 20-21. Conclusion Grant, Bruce. 1999. Fine tuning the peppered moth. Evolution 53: 980—984. Kettlewell's data are easily accounted for by die unsurprising . 2002. Sour grapes of wrath. Science 297: 940-941. Feinstein, A.R. 2002. Principles of Medical Statistics. London: Chapman and fact that you can recapture more moths when you release Hall. p. 265. more—that and normal experimental variation. When the Forrest, Barbara, and Paul R. Gross. 2004. Creationism's Trojan Horse: The effect of moonlight is included in die calculation, the calcu- Wedge of Intelligent Design. New York: Oxford University Press. lated curve fits even closer to Kettlewell's data. We have no Hooper, Judith. 2002. Of Moths and Men: The Untold Story of Science and the Peppered Moth. New York: W.W. Norton. need of Hooper's perverse, ad hoc hypothesis of fraud. ISO. 1993. Guide to the Expression of Uncertainly in Measurement. Geneva: Hooper's claims are moonshine; diey are based on a lack of International Organization for Standardization. understanding of Kettlewell's experiments in particular and Kettlewell, H.B.D. 1955. Selection experiments on industrial melanism in the Lepidoptera. Heredity 9: 323-342. experimental science in general. Hooper evidently did not . 1956. Further selection experiments on industrial melanism in the realize that the change in recapture numbers began before lepidoptera. Heredity 10 (Pan 3): 287-301. Kettlewell could have read die letter that supposedly triggered -. 1959. Darwin's missing evidence- Scientific American 200:48-53. diis change, let alone consider die most likely cause of die Majerus, M.E.N. 1998. Melanism: Evolution in Action. Oxford: Oxford University Press. Chapter 6. changes she saw, exposure to moonlight. Hooper and Sargent . 2002. Moths. London: HarperCollins. Chapter 9. should have performed a careful analysis before Hooper's pre- Majerus, M.E.N, CFA. Brunton. and J. Stalker. 2000. A bird's eye view of sumptuous insinuation of fraud. the peppered moth. Journal of Evolutionary Biology 13:155—159. Mallet, Jim. 2004. The peppered moth: A black and white story after all. Kettlewell's conclusion—that predadon by birds was a major Genetics Society News. 50: 34-38; also available online at factor in promoting industrial melanism—was based on at least www.genctics.org.uk/?page=issue_50. four lines of inquiry, as detailed above. It did not rely on the Musgrave. Ian. 2005. Paint it black: The peppered moth story. In Icons of release-recapture experiments alone. It is also supported by at least Evolution, Society for the Study of Evolution Education Committee, to be published. diirty studies of different modi species diat also developed melanic Shapiro. Arthur M. 2002. Paint it blade Evolution 56: 1885-1886. •

2 8 March/April 2005 SKEPTICAL INQUIRER