For my books, click here.
(This list excludes keywords and addresses.)
T P Hutchinson: List of papers (from 1993, summaries are given)
---------------------------------------------------------------
1. Hutchinson, T P 1972. Delay at a fixed time traffic signal.
II: Numerical comparisons of some theoretical expressions.
Transportation Science, 6, 286-305.
2. Hutchinson, T P, and Satterthwaite, S P 1974. Analysis of some
characteristics of road traffic accidents by use of police reports.
6th Conference of the Universities Transport Study Group, held at
the University of Manchester Institute of Science and Technology.
3. Hutchinson, T P 1974. Urban traffic speeds. II: Relation of the
parameters of two simpler models to size of city and time of day.
Transportation Science, 8, 50-57.
4. Hutchinson, T P 1974. The distribution of traffic in towns:
Further validation of Vaughan's model by correlating its
parameters with size of city and time of day. Traffic Engineering
and Control, 15, 770-771.
5. Hutchinson, T P 1974. The communication of transport research.
Traffic Engineering and Control, 15, 775.
6. Hutchinson, T P 1974. The control of right-turning vehicles at
signal-controlled intersections: A comment on a suggestion by
Al-Salman and Salter. Traffic Engineering and Control, 15, 920-923.
7. Hutchinson, T P 1975. Factors affecting the times till death of
pedestrians killed in road accidents. Injury: the British Journal of
Accident Surgery, 6, 208-212.
8. Hutchinson, T P 1975. Witnesses' estimates of the speeds of traffic
accidents. Accident Analysis and Prevention, 7, 27-35.
9. Goodwin, P B, Hutchinson, T P, and Wright, C C 1975. The perception
of vehicle speeds by pedestrians. Zeitschrift fur Verkehrssicherheit,
21, 13-18.
10. Jolliffe, J K, and Hutchinson, T P 1975. A behavioural explanation
of the association between bus and passenger arrival times at a bus
stop. Transportation Science, 9, 248-282.
11. Hutchinson, T P 1975. Factors affecting the severity of injury to
adult pedestrians involved in road accidents. 5th International
Conference of the International Association for Accident and Traffic
Medicine, held in London.
12. Hutchinson, T P, and Jones, I S 1975. The separation of the effects
of driver and of vehicle on type of accident. 5th International
Conference of the International Association for Accident and Traffic
Medicine, held in London.
13. Hutchinson, T P 1976. Statistical aspects of injury severity. Part I:
Comparison of two populations when there are several grades of
injury. Transportation Science, 10, 269-284.
14. Hutchinson, T P 1976. Statistical aspects of injury severity. Part II:
The case of several populations but only three grades of injury.
Transportation Science, 10, 285-299.
15. Hutchinson, T P 1976. Combining two-tailed rank-correlation
statistics. Applied Statistics (Journal of the Royal Statistical
Society, Series C), 25, 21-25.
16. Goodwin, P B, and Hutchinson, T P 1977. The risk of walking.
Transportation, 6, 217-230.
17. Hutchinson, T P 1977. Application of Kendall's partial tau to a
problem in accident analysis. International Journal of
Bio-Medical Computing, 8, 277-281.
18. Hutchinson, T P 1977. Intra-accident correlations of driver injury
and their application to the effect of mass ratio on injury
severity. Accident Analysis and Prevention, 9, 217-227.
19. Hutchinson, T P 1977. Latent structure models applied to the joint
distribution of drivers' injuries in road accidents. Statistica
Neerlandica, 31, 105-111.
20. Hutchinson, T P 1977. Universities Transport Study Group:
A Conference report. Traffic Engineering and Control, 18, 211.
21. Hutchinson, T P, and Mayne, A J 1977. The year-to-year variability
in the numbers of road accidents. Traffic Engineering and Control,
18, 432-433.
22. Hutchinson, T P 1977. On the relevance of signal detection theory
to the correction for guessing. Contemporary Educational Psychology,
2, 50-54.
23. Hutchinson, T P, and Satterthwaite, S P 1977. Mathematical models for
describing the clustering of sociopathy and hysteria in families:
A comment on the recent paper by Cloninger et al. British Journal of
Psychiatry, 130, 294-297.
24. Hutchinson, T P 1977. The method of m rankings when the numbers of
observations in each cell are not all unity. Computers and Biomedical
Research, 10, 345-361.
25. Hutchinson, T P 1978. An extension of the signal detection model of
information retrieval. Journal of Documentation, 34, 51-54.
26. Satterthwaite, S P, and Hutchinson, T P 1978. A generalisation of
Gumbel's bivariate logistic distribution. Metrika, 25, 163-170.
27. Hutchinson, T P 1978. Some statistical methods useful in the
analysis of road accident data. 1st Course on Crashworthiness in
Transportation Systems, held at the Ettore Maiorana Centre for
Scientific Culture, Erice, Sicily.
28. Hutchinson, T P, and Harris, R A 1978. Recent trends in traffic
injury. Injury: the British Journal of Accident Surgery, 10,
133-137 (Annotations, 137-138).
29. Grime, G, and Hutchinson, T P 1979. Vehicle mass and driver
injury. Ergonomics, 22, 93-104.
30. Hutchinson, T P 1979. On the relative frequencies of collisions
between vehicles of different masses. Zeitschrift fur
Verkehrssicherheit, 29, 61-63.
31. Hutchinson, T P 1979. A comment on replicated paired comparisons.
Applied Statistics (Journal of the Royal Statistical Society,
Series C), 28, 163-169.
32. Hutchinson, T P 1979. The validity of the chi-squared test when
expected frequencies are small: A list of recent research
references. Communications in Statistics, Part A: Theory and
Methods, A8, 327-335.
33. Hutchinson, T P 1979. Four applications of a bivariate Pareto
distribution. Biometrical Journal (Biometrische Zeitschrift),
21, 553-563.
34. Hutchinson, T P 1980. An easy method of calculating approximate
recurrence risks using a multifactorial model of disease
transmission. Annals of Human Genetics, 43, 285-293.
35. Hutchinson, T P 1980. The definition of restraint effectiveness.
Accident Analysis and Prevention, 12, 81-93.
36. Hutchinson, T P 1980. Approximating a trivariate normal probability
that is of special relevance to the multifactorial model of disease
transmission. Annals of Human Genetics, 44, 107-111.
37. Hutchinson, T P 1980. An assessment of the usefulness of route advice
given by British Rail to passengers. Journal of Consumer Studies and
Home Economics, 4, 135-150.
38. Hutchinson, T P 1980. Partial knowledge and the theoretical basis
of linear corrections for guessing. Contemporary Educational
Psychology, 5, 227-231.
39. Hutchinson, T P, and Lai, P W 1980. Experience with the use of the
program CATLIN in analysing categorical data, with examples from
road accident studies. In M M Barritt and D Wishart (Editors),
COMPSTAT 1980. Proceedings in Computational Statistics, pp. 515-522.
Wien: Physica-Verlag.
40. Zlotnicki, J, Hutchinson, T P, and Kendall, D L 1980. Some problems
and prospects with commercial vehicle safety, illustrated by case
reports of accidents involving ergonomic factors. In D J Oborne and
J A Levis (Editors), Human Factors in Transport Research. Volume 1,
pp. 151-159. London: Academic Press.
41. Hutchinson, T P, and Haslegrave, C M 1980. Determination of
patterns of human body measurements by use of partial correlations.
Ergonomics, 23, 475-483.
42. Hutchinson, T P 1981. A review of some unusual applications of
signal detection theory. Quality and Quantity, 15, 71-98.
43. Hutchinson, T P 1981. Causes of death in road crashes: Evidence from
the routinely-published statistics of several countries. American
Association for Automotive Medicine Quarterly/Journal, 3(1), 39-40.
44. Hutchinson, T P 1981. An assessment of the information given in
railway timetable-leaflets. Journal of Consumer Studies and Home
Economics, 5, 239-246.
45. Hutchinson, T P 1981. Compound gamma bivariate distributions. Metrika,
28, 263-271.
46. Hutchinson, T P 1981. Disagreements when using ordered categories.
Quality and Quantity, 15, 593-596.
47. Hutchinson, T P, and Adams, V 1981. International statistics of road
fatalities. Transport Reviews, 1, 393-397.
48. Hutchinson, T P, and Lai, P W 1981. Statistical aspects of injury
severity. Part III: Making allowance for differences in the
assessment of level of trauma. Transportation Science, 15, 297-305.
49. Hutchinson, T P 1982. Statistical aspects of injury severity.
Part IV: Matched data. Transportation Science, 16, 83-105.
50. Hutchinson, T P 1982. Signal probability and the slope of the ROC:
A comment on Treisman. Psychological Bulletin, 91, 193-197.
51. Frary, R B, and Hutchinson, T P 1982. Willingness to answer
multiple-choice questions as manifested both in genuine and in
nonsense items. Educational and Psychological Measurement, 42,
815-821.
52. Hutchinson, T P 1982. Some theories of performance in multiple-choice
tests, and their implications for variants of the task. British
Journal of Mathematical and Statistical Psychology, 35, 71-89.
53. Hutchinson, T P 1982. Disease concordance and sex similarity in
twins: Application of a multifactorial model with latent structure.
Journal of Epidemiology and Community Health, 36, 155-156.
54. Grime, G, and Hutchinson, T P 1982. The influence of vehicle weight
on the risk of injury to drivers. Ninth International Technical
Conference on Experimental Safety Vehicles, held in Kyoto,
pp. 726-741. Washington, D.C.: National Highway Traffic Safety
Administration.
55. Hutchinson, T P 1983. A bivariate normal model for intra-accident
correlations of driver injury, with application to the effect of mass
ratio. Accident Analysis and Prevention, 15, 215-224.
56. Hutchinson, T P 1983. A note on applications of the competing risks
model. Accident Analysis and Prevention, 15, 225-226.
57. Hutchinson, T P, and Roe, M S 1983. Latent structure modelling of
trip distribution. Compendium of Technical Papers, 53rd Annual
Meeting of the Institute of Transportation Engineers, held in London,
pp. 11.22-11.24. Washington, D.C.: Institute of Transportation
Engineers.
58. Hutchinson, T P, Allen, A M, Gornell, A W, and Liew, V K 1983.
Judging the centres of irregular shapes: How much variability is
there? Ergonomics, 26, 981-984.
59. Hutchinson, T P, and Nicholl, J P 1983. Recommendations about
collecting bus headway data and estimating expected waiting times.
Traffic Engineering and Control, 24, 529-531.
60. Hutchinson, T P 1984. Medical statistics on road accident injury in
several countries. In S Yagar (Editor), Transport Risk Assessment,
pp. 43-76. Waterloo: University of Waterloo Press. An earlier
version appeared in Covjek i Promet, 9, 17-30 (1983), under the
title "Deaths and injuries in road accidents: Evidence from the
mortality and health statistics of several countries" (with
"health" erroneously omitted from the printed title).
61. Hutchinson, T P 1984. Using the bivariate normal distribution to
construct probability models in the health sciences. Communications
is Statistics - Theory and Methods, 13, 1723-1733.
62. Hutchinson, T P 1984. Risk in transport. A symposium report. Traffic
Engineering and Control, 25, 330-331.
63. Hutchinson, T P 1984. Cross-national comparison of the routine
collection of data on the nature and severity of injury in road
crashes. Presented at the International Workshop on the Methodology
of Modelling Road Accident and Injury Patterns, organised by the
International Drivers' Behaviour Research Association, held at the
University of Sussex.
64. Hutchinson, T P 1984. Nonsense items in multiple-choice tests.
Presented at the London Conference of the British Psychological
Society, held at the City University. (ERIC Document ED254537,
Educational Resources Information Center.)
65. Hutchinson, T P 1985. Analysing severity data when assessors
differ in their usage of the categories. The Statistician, 34,
183-195. (Presented at the Conference on Statistics in Health,
organised by the Institute of Statisticians, held at the
University of Kent, July 1984.)
66. Hutchinson, T P 1985. Reliability of motor vehicle fatality statistics:
An international perspective. Canadian Journal of Public Health, 76,
413-414.
67. Hutchinson, T P 1985. Predicting performance in variants of the
multiple-choice test. Presented at the Fourth European Meeting of the
Psychometric Society and the Classification Societies, held at the
University of Cambridge. (ERIC Document ED263177, Educational Resources
Information Center.)
68. Hutchinson, T P 1985. Presenting one probability distribution as a
function of another - some applications. American Journal of
Mathematical and Management Sciences, 5, 103-123.
69. Hutchinson, T P 1986. Statistical modelling of injury severity,
with special reference to driver and front seat passenger in
single-vehicle crashes. Accident Analysis and Prevention, 18, 157-167.
(Presented at the International Workshop on the Methodology of
Modelling Road Accident and Injury Patterns, organised by the
International Drivers' Behaviour Research Association, held at the
University of Sussex, July 1984.)
70. Hutchinson, T P, Burt, N, Cuzner, A, and Howell, R 1986. Three
categories of urban bus passengers. Highways and Transportation,
33(10), 14-16.
71. Hutchinson, T P 1986. Evidence about partial information from an
answer-until-correct administration of a test of spatial
reasoning. Contemporary Educational Psychology, 11, 264-275.
72. Hutchinson, T P, and Rowell, S 1986. Points systems for car
insurance. Insurance: Mathematics and Economics, 5, 255-259.
73. Hutchinson, T P, and Alderson, J C 1986. Routine road accident data:
Publications and their indexing. In J O Asalor, E A Onibere, and
G C Ovuworie (Editors), Road Traffic Accidents in Developing
Countries - Volume 1, pp. 461-488. Lagos: Joja Educational Research
and Publishers.
74. Hutchinson, T P, and Barton, D C 1987. A mechanical reasoning test
with answer-until-correct directions confirms a quantitative
description of partial information. Research in Science and
Technological Education, 5, 93-101.
75. Hutchinson, T P, and Tang, K Y 1987. Letter on "The value of latent
class analysis in medical diagnosis" by Rindskopf and Rindskopf.
Statistics in Medicine, 6, 529-530.
76. Hutchinson, T P 1987. Re "Analytical potential for multiple
cause-of-death data". American Journal of Epidemiology, 126, 158-159.
77. Hutchinson, T P 1988. Inter-observer agreement about traffic
conflicts: A fourth opinion. Traffic Engineering and Control, 29,
235-237.
78. Hutchinson, T P 1992. Randomization tests and the generalised
method of m rankings. Computer Methods and Programs in
Biomedicine, 37, 163-164.
79. Hutchinson, T P 1992. In the multifactorial model of disease
transmission, why is the rank correlation sensitive to choice of
bivariate distribution? Annals of Human Genetics, 56, 159-161.
80. Hutchinson, T P 1992. Self-indexing. The Indexer, 18, 105-106.
81. Hutchinson, T P, and Lai, C D 1992. Concepts of stochastic ageing -
Who cares? Safety and Reliability, 12(3), 7-12.
82. Hutchinson, T P 1992. Providing information to the traveller by
public transport. Presented at the 14th Conference of Australian
Institutes of Transport Research, held in Sydney.
83. Hutchinson, T P 1992. Discussion of road accidents in Iran.
Journal of Engineering, Islamic Republic of Iran, 5, 75-76.
84. Hutchinson, T P 1992. What do they have in common? Royal
Statistical Society News and Notes, 19(1), 7.
This is a brief letter on who are the people who most
frequently publish comments on other statistical papers; and
some statistical features of such publications - the most
productive fraction p of commentators account for
approximately a fraction sqrt{p} of papers commented upon.
85. Hutchinson, T P 1993. On macroscopic description of urban
traffic speeds. Journal of the Operational Research Society,
44, 209-210.
A V Hill and W C Benton (J Opl Res Soc, 43, 1992, 343-351)
reported on vehicle scheduling problems in which vehicle speeds
depend on geographical location and time of day. They adopted
a broad brush approach, and approximated the travel speed from
i to j, denoted rij, by the average of speeds associated with
the origin and the destination, (ri + rj)/2. The present
contribution calls attention to work in which vehicle speed in
a city is approximated by a function of the distance from the
city centre. There is both theoretical and empirical work of
this kind.
86. Hutchinson, T P 1993. A guide to bivariate ideas of quantal
response analysis, as applied in biometrics and econometrics.
Biometrical Journal, 35, 99-108.
A review is provided of the several bivariate generalisations
of quantal response analysis that have appeared in the biometric
and econometric literatures. There are three main types:
(i) where a binary outcome is the result of two stimulants, and
thus the bivariate distribution of the thresholds for response
is relevant; (ii) where three or more alternative outcomes may
arise from a single stimulant; and (iii) where the response
itself is bivariate (i.e., two types of response may
simultaneously be observed).
86. Hutchinson, T P 1993. The seventh-root formula for a trivariate
normal probability. The American Statistician, 47, 102-103.
A simple method of approximating the trivariate normal
integral in terms of the bivariate marginal probabilities is
proposed. It is applied to the multifactorial model of disease
inheritance.
88. Hutchinson, T P, and Lai, C D 1993. As regards reliability, what
is in the Encyclopedia of Statistical Sciences? Safety and
Reliability, 13(1), 13-21.
The Encyclopedia of Statistical Sciences was published in
nine volumes plus a supplementary volume over the years 1982 to
1989. In this paper, a brief guide is given to those of the
substantial entries concerning statistical reliability which
may be of relevance to engineers and reliability analysts.
89. Hutchinson, T P 1993. Kappa muddles together two sources of
disagreement: Tetrachoric correlation is preferable. Research in
Nursing and Health, 16, 313-316.
When assessing agreement between experts, it is important to
distinguish between disagreements that can and those that cannot
be explained by different placing of the boundaries between
categories. Cohen's kappa statistic is affected by both types of
disagreement, tetrachoric correlation only by the second.
90. Hutchinson, T P 1993. Second attempts at multiple-choice test
items. Journal of Statistical Computation and Simulation, 47,
108-112.
Sometimes, an examinee answering a multiple-choice item is
permitted a second attempt when the first response is wrong.
Presumably, the probability of being correct at the second
attempt (c2) ought to be predictable from the probability of
success at the first attempt (c1). But (a) what kind of theory
predicts a relatively low c2 for given c1, and (b) what kind of
theory predicts a relatively high c2 for given c1? This note
conjectures the answers are as follows: (a) some correct
options are known to be correct, but no distractors are known
to be distractors, and (b) no correct options are known to be
correct, but distractors are sometimes recognised as such.
It is hoped that some reader can supply a neat proof.
91. Hutchinson, T P 1993. Comments on the multivariate normal integral.
Journal of Statistical Computation and Simulation, 47, 112-114.
In the statistical literature, renewed interest in calculating
the normal integral in three and more dimensions is evident.
This notes draws attention to three applications areas where
multivariate normal probabilities are often required, and which
have developed their own literatures on this. These areas are
heredity (the multifactorial model of disease transmission),
structural failure, and econometric choice theory. A few
references are provided, as a way into each of these.
92. Hutchinson, T P 1993. Discussion of data on the killing by cold of
fruit fly larvae in mangosteens. The Kasetsart Journal: Natural
Sciences, 27, 226-229.
I Burikam and colleagues (The Kasetsart Journal: Natural
Sciences, 25, 1991, 251-255) have recently quantified the
relationship between exposure to cold and death of fruit fly
larvae infesting mangosteens. It is shown here that the results
are robust with respect to a number of minor changes that could
have been made in the statistical analysis. But the results are
very sensitive to the choice of statistical model assumed
(probit or logit) - whereas .0032% of observation lie more than
4 standard deviations above the mean of a normal distribution,
about 20 times than number (.071%) lie more than 4 standard
deviations above the mean of a logistic distribution.
93. Hutchinson, T P 1994. On overconfidence in multiple-choice tests.
Psychological Record, 44, 253-255.
D Zakay and J Glicksohn (Psychological Record, 42, 1992,
519-524) found a negative relationship between overconfidence and
performance in a multiple-choice test. The present paper argues
that, contrary to what they say, the result could be artificial.
94. Hutchinson, T P 1994. Finding corrigenda to journal articles.
Australian and New Zealand Journal of Serials Librarianship,
4(4), 69-72.
A description is given of compiling a book that lists
corrections, addenda, and comments that were published in
any of 78 statistics journals over the period 1970-1991. The
number of articles corrected, added to, or commented on was
some 3200. A similar but much shorter index listing 276 articles
in 9 psychometric journals has also been published.
95. Hutchinson, T P 1994. Models for responses to multiple-choice
items. Presented at the 12th Australian Statistical Society
Conference, held at Monash University.
Some models are proposed for how examinees respond to questions
in multiple-choice tests. Their predictions about how
second-attempt performance is related to first-attempt performance
are derived. (i) The starting point is the model in which
knowledge is all-or-nothing: either the correct answer is known
(and correctly chosen), or the probability of a correct response
is at the chance level. A feature of this model is that it makes
no provision for partial information: for example, when a second
attempt is permitted at questions initially answered wrongly, the
model predicts performance to be at the chance level, but this
contrasts with the empirical finding of above-chance performance
in this situation. (ii) The next idea to be considered is that the
examinee may be able to eliminate one or more of the distractors
as being definitely wrong, and then he or she guesses at random
from among the remaining options. The prediction of this model for
the relationship between first- and second-attempt performance is
derived. (iii) Putting the above two models together, we can
imagine one in which there is some probability of the correct
answer being recognised as correct, and some probability of a
distractor being recognised as wrong; and, for simplicity, one
might take these probabilities to be equal. The prediction for the
relationship between first- and second-attempt performance is
derived. (iv) A few other models are also proposed, including an
asymmetric model, one that incorporates careless slips, and one
that incorporates misinformation.
96. Hutchinson, T P 1994. The greatest new development in traffic
engineering since the traffic signal? Road and Transport Research,
3(4), 91-94. Reprinted in Traffic Engineering and Control, 36,
156-157 (1995).
A triple conflict between traffic streams - for instance, an
appreciable right-turning flow, as well as heavy straight-through
flows - is often the key factor leading to limited capacity and
high delay at a junction, either three-legged T or four-legged
crossroads. It is, however, possible to design a junction so that
such a conflict is replaced by three lesser conflicts, each between
only a pair of flows. This leads to higher capacity and less delay,
for a given amount of road space. Such designs have been publicised
by R Goldblatt, F Mier, and J Friedman (ITE Journal, 64, 1994,
35-42). The present contribution draws attention to support given
by T P Hutchinson's article in Traffic Engineering and Control,
15, 1974, 920-923; that support was based on queueing formulae and
numerical calculations using them.
97. Hutchinson, T P 1995. Asking sensitive questions in surveys.
(Classroom Note.) Teaching Statistics, 17, 43.
98. Hutchinson, T P 1996. Waiting for buses: Size-weighted means.
(Classroom Note.) Teaching Statistics, 18, 9.
99. Hutchinson, T P 1996. Performance on items in conventional
multiple-choice and in multiple true-false formats. Educational
Research Quarterly, 19(3), 3-7.
It is shown how performance on items administered in multiple
true-false format may be predicted from performance when the items
are administered in conventional multiple-choice format. (The theory
is essentially that described by T P Hutchinson in Br J Mathl Statl
Psychol, 1982, 35, 71-89.)
100. Hutchinson, T P 1996. Nonunique estimates of ability. Journal of
Statistical Planning and Inference, 55, 262-264.
For some models, it can happen that datasets arise for which
the likelihood function has more than one maximum. That is bad
enough when the parameter just refers to some statistical
distribution, but it can also happen with a person parameter
in item response theory (IRT), and be important in determining
the person's educational and career opportunities. (The IRT
model discussed here is the three-parameter logistic.) The
problem has practical and philosophical sides to it, as follows.
When computing estimates of ability, how is it known that the
global maximum of likelihood has been found? If the likelihoods
at two maxima are very similar, how is one ability or the other
assigned to the examinee?
101. Hutchinson, T P 1996. On the generalised Friedman test.
Computational Statistics and Data Analysis (Statistical
Software Newsletter section), 21, 473-476.
Nonparametric methods for two-way analysis of variance were
discussed by V W Rahlfs and H Zimmermann (Computational
Statistics and Data Analysis, 20, 1995, 101-110). Several points
stemming from their paper are made in the present contribution.
The most important are: (a) chi-squared can be a poor
approximation to the null distribution of the test statistic,
and so a randomisation test is desirable; (b) as has been
argued by J de Kroon and P van der Laan (Statistica Neerlandica,
37, 1983, 1-14) and others, allowance needs to be made for
differing numbers of observations per row.
102. Hutchinson, T P 1997. Mismatch models that permit partial
information to be shown. In W J van der Linden and R K Hambleton
(Editors), Handbook of Modern Item Response Theory, pp. 481-494.
New York: Springer.
Conventional IRT (item response theory) models make no allowance
for the examinee having partial information about the question
asked. The models are unable to predict what might happen in
situations which allow the partial information to be shown.
Relations between probabilities of correctness in different formats
of test - for example, with different numbers of options to choose
from, or permitting a second attempt at items initially answered
wrongly - do not fall within their scope. The present chapter puts
forward what it terms mismatch theory, which overcomes this
limitation. It is in the style of signal detection theory.
103. Hutchinson, T P 1997. Improving rater agreement studies. American
Journal of Roentgenology, 168, 1382.
Studies of agreements between raters are important in
medicine and elsewhere. Several improvements to the reporting of
such studies are suggested here. Among these are (a) be clear
about the process of selection of the objects that were judged,
and (b) do not use Cohen's kappa, but instead calculate
separate measures of the correlation between raters and of their
relative bias.
104. Hutchinson, T P 1997. Comments on McQuay et al., Pain, 64 (1995)
331-335. Pain, 73, 107-108.
It is sometimes claimed that in studies where the average
response to treatment is large, the average response to placebo
also tends to be large. The purpose of this note is to propose
a possible mechanism by which such a correlation may occur.
This is that it is a consequence of the treatment and placebo
groups in a given study being more similar to each other (in
respect of various situational factors and patient
characteristics) than they are to the groups in other studies.
This idea is applied to data in an article by H McQuay,
D Carroll, and A Moore (Pain, 64, 1995, 331-335). The question
is raised whether, given that the mean placebo response varies
across studies (for unexplained reasons) from 11 to 29 (the
amount that McQuay et al. noted), the treatment-placebo
correlation might be as large as 0.87 (which is what McQuay
et al. found). The answer is that this correlation is larger
than expected if the proposed mechanism is correct, but by usual
criteria is not unreasonably large.
105. Hutchinson, T P 1997. Comment on correlation in skew distributions.
Journal of General Psychology, 124, 211-215.
If the marginals of a bivariate normal distribution are
transformed (e.g., by exponentiation), the product-moment
correlation is reduced. Dunlap et al. (Journal of General
Psychology, 122, 1995, 365-377) recently discussed this effect in a
psychological context. The present paper draws attention to related
work in physical sciences (such as structural mechanics, geology,
and meteorology).
106. Hutchinson, T P 1997. Radioactivity half-lives considered as data.
Journal of Applied Mathematics and Decision Sciences, 1, 67-71.
Half-lives of radioactive nuclides range over more than 20
orders of magnitude. It is striking that, nevertheless, statistical
laws may be discovered in these numbers: a log-normal distribution
provides a good description. This is shown using the cumulative
distribution. The final paragraph of the paper observes that
half-lives and disintegration energies are thought of as constants,
not as realisations of some random process, and it would therefore
be misguided to ask why they should exhibit particular statistical
features, or to imagine a stochastic mechanism by which they are
generated.
107. Hutchinson, T P 1997. The binomial and Poisson distributions.
(Classroom Note.) Teaching Statistics, 19, 68.
In their concern to distinguish the Poisson distribution
from the binomial, some books neglect the similarities between
these distributions. Both use the ideas of a constant probability
of an event occurring, and of independence between all the
opportunities for the event to occur. An example is given,
in which road traffic passing a point can be modelled using
either distribution.
108. Myors, B, and Hutchinson, T P 1997. Using simulation to infer
the structure of vocational interests. In Proceedings of the
39th Annual Conference of the International Military Testing
Association, pp. 323-327. Canberra: Defence Publishing Service.
Vocational assessment methods have been used for
counselling workers in civilian and military roles for many
years. This paper proposes new methods of statistical
inference applicable to the structure of vocational interests.
Our approach to inference in this area involves
variables-in-common models constructed in terms of Holland's
widely adopted theory. We use the models to test varying forms
of hypothesised underlying structure. One test examines the
hexagonal arrangement proposed by Holland, another extends this
in the direction of Gati's ideas. The statistical testing is not
based upon algebra, but upon simulation of raw data according to
the models. The technique is illustrated using a military dataset.
109. Hutchinson, T P 1997. A heteroscedastic bivariate distribution
arising from a model for rater agreement, and its fitting by
simulation. Computational Statistics, 12, 497-503.
The starting point is a dataset showing that two raters,
judging proficiency in spoken Russian, appear to disagree more
about the relatively expert speakers than about the novices.
A bivariate distribution is invented and shown to fit the data
better than the bivariate normal distribution does. The chief
features of the distribution are that it is a variables-in-common
model, true score plus error for each rater, and that the scatter
of error is greater when the true score is high than when it is
low. The method of fitting the distribution to the data is
simulation. Accordingly, an explicit expression for the joint
distribution of the two observed scores is not required. The
software used has ranking and recoding commands of one line each,
so it is easy to ensure the fitted marginal distributions exactly
match the data, and it is unnecessary to estimate parameters
representing the boundaries between the grades of rating.
110. Hutchinson, T P 1997. What effect does an untreated aneurysm
have on life expectancy? Canadian Journal of Neurological
Sciences, 24, 357-358.
If a patient is found to have a cerebral aneurysm that is
asymptomatic, should the surgeon operate or leave well alone?
The present contribution discusses some aspects of a paper by
R Leblanc and K J Worsley (Can J Neurol Sci, 22, 1995, 30-35).
A formula is put forward that gives the expected lifetime lost
when an extra risk (e.g., aneurysm rupture) that is constant
through time is added to all the existing risks. The formula
is expressed in terms of the level of this extra risk, along
with the mean and standard deviation that lifetimes would have
if that risk were absent.
111. Hutchinson, T P 1998. On a recognition task in which some
distractors were half-familiar. International Journal of Aging
and Human Development, 46, 21-24.
Probabilistic models are suggested for the task of recognising
word-pairs, where the distractors may be pairs of new words, or
may be a new word paired with a previously-seen word. These
models are relevant to an experiment recently reported by
M Isingrini et al. (International Journal of Aging and Human
Development, 41, 1995, 79-88), and suggest rather different
conclusions to theirs - namely, that the elderly differ from the
young in both learning and response-selection characteristics.
112. Hutchinson, T P, and Cairns, D 1998. Difference or ratio:
A note on two-treatment, four-sequence analysis. Biometrics,
54, 788-789.
In a two-treatment four-sequence experiment, patients are
randomised to one of the four sequences DD, DP, PD, or PP (D=drug,
P=placebo). Elswick and Uthoff (Biometrics, 1989) proposed a
"nonparametric" method for analysing such data. This wording may
suggest that monotonic but nonlinear transformations have no
effect. However, the nonparametric aspect of Elswick and Uthoff's
methods only enters after the initial calculations - which
consist of contrasting the results in the second period with those
in the first - have been performed. Specifically, if the method of
contrasting consists of taking ratios of the raw numbers, rather
than taking differences, the conclusions for Elswick and Uthoff's
example are changed. It is also shown how the software StatXact 3
may be used in this context.
113. Lai, C-D, Rayner, J C W, and Hutchinson, T P 1998. Properties
of the sample correlation of the bivariate lognormal distribution.
In L Pereira-Mendoza, L S Kea, T W Kee, and W-K Wong (Editors),
Statistical Education - Expanding the Network: Proceedings of the
Fifth International Conference on Teaching of Statistics, Volume 1,
pp. 309-315. Voorburg: International Statistical Institute.
Most statistics students know that the sample correlation
coefficient may be used to estimate the population correlation
coefficient. If the parent population is bivariate normal, this
does not cause any trouble. However, if the marginals are
nonnormal, the estimated value from a sample may be quite
different from the population value. Our simulations indicate
that for bivariate lognormal distributions that are highly
skew, the bias in estimating rho can be very large, and is
substantially reduced only with millions of observations. This
example could serve as an exercise for statistics students to
realise some of the pitfalls in using the sample correlation to
estimate the population correlation.
114. Hutchinson, T P 1998. Mean or median for exponential data?
(Classroom Note.) Teaching Statistics, 20(3), 92-93.
For a skewed population, it is often considered that the
median captures the idea of "typical value" better than the mean
does. But what method should be used to estimate the population
median? It seems natural to use the sample median for this.
However, suppose one believes that the parent distribution is
the exponential. Then it is advantageous to estimate the
population median by first calculating the sample mean (and then
multiplying by 0.693).
115. Chekaluk, E, Hutchinson, T P, and Cairns, D 1998. Repeated
measures ANOVA for responses developing over time. European
Journal of Anaesthesiology, 15, 381-382.
The starting point is a paper by K Abt (European Journal of
Anaesthesiology, 13, 1996, 427-431). Comment is made on methods
of comparing the responses at several points in time in two
groups of patients. The technique of repeated measures analysis
of variance, and how this is handled in SPSS, is described.
116. Hutchinson, T P, and Myors, B 1998. Comparison of the structures
of vocational interests of men and women. In Proceedings of the
1st International Work Psychology Conference. Sheffield:
University of Sheffield, Institute of Work Psychology
(ISBN 0 9533504 0 1).
Anderson et al. (Journal of Vocational Behavior, 50, 1997,
349-364) reviewed evidence on whether there are gender differences
in Holland's RIASEC model of vocational interests, and
re-published seven pairs of RIASEC correlation matrices (one for
women and one for men, from seven studies). We have re-examined
these fourteen correlation matrices, and some of our conclusions
are different from those of Anderson et al. Specifically, we have
detected asymmetries that are consistent across studies for each
gender separately, but which are different for women and men. For
women, correlations involving I or R tend to be large, and
correlations involving A tend to be small. For men, correlations
involving S or C tend to be large, and correlations involving A
tend to be small.
117. Hutchinson, T P 1998. Note on probability distributions for
generation time. Journal of Applied Microbiology, 85, 192-193.
Ratkowsky et al. (Journal of Applied Bacteriology, 80, 1996,
131-137) reported on the variation in generation times of a
Pseudomonas fluorescens culture. They advocated the use of the
gamma distribution to describe this. However, it is argued in the
present contribution that the main message of their paper is that
the coefficient of variation (ratio of standard deviation to mean)
does not depend upon temperature; what is presented tells us
nothing about the shape of the distribution. (The shape in
Figure 2 of Ratkowsky et al. is not the shape of the data, but is
determined by the coefficient of variation in the data, along with
the assumption of the distribution being gamma.)
118. Hutchinson, T P 1998. Two aspects of reliability. Australian
Journal of Osteopathy, 9(2), 5.
This is a short letter drawing readers' attention to the
distinction between "Do the ratings mean the same thing to the
two experts making them?" and "Do the experts place the patients
in the same order?"
119. Hutchinson, T P 1999. Measuring the congruence of worker and
workplace: A correlational approach. Australian Journal of
Career Development, 8(1), 18-20.
Suppose we know the Holland three-letter codes for a worker
and for a workplace - for example, somebody assessed to be RCI
may be working in an occupation considered to be CRS. This paper
considers how to measure the similarity (or congruence) between
the codes. Previous methods are briefly reviewed. Then it is shown
how the familiar idea of correlation may be adapted to the
problem. Some of the existing methods are shown to be special
cases of the new approach.
120. Lai, C D, Rayner, J C W, and Hutchinson, T P 1999. Robustness of
of the sample correlation - The bivariate lognormal case. Journal
of Applied Mathematics and Decision Sciences, 3, 7-19.
The sample correlation coefficient R is almost universally
used to estimate the population coefficient rho. If the pair
(X,Y) has a bivariate normal distribution, this does not cause
any trouble. However, if the marginals are nonnormal, particularly
if they have high skewness and kurtosis, the estimated value from
a sample may be quite different from the population correlation
coefficient rho. The bivariate lognormal is chosen for this
robustness study. Two approaches are used: simulation and algebra.
Our simulation indicates that for the bivariate lognormal, the
bias in estimating rho can be very large if rho is nonzero, and it
can be substantially reduced only after a large number (millions)
of observations. This phenomenon, though unexpected at first, was
found to be consistent with our algebraic numerical analysis.
121. Hutchinson, T P 1999. Familial association of disease and the
structure of trivariate distributions. Annals of Human Genetics,
63, 539-544.
In its usual form, the multifactorial model of disease
transmission assumes that the liabilities to disease have a
multivariate normal distribution. This paper studies how
sensitive to this assumption are the quantitative results from
the model. Accordingly, bounds are established for the
probability of a child having a disease, given that both
parents have it and taking the heritability of the disease to be
known. Unfortunately, these bounds turn out to be wide. For
example, a probability that is 0.38 under the trivariate normal
model may be as low as 0.12 or as high as 0.78 under other
trivariate models, even if attention is restricted to those of
variables-in-common form. The broader statistical issue of the
meaning of trivariate dependence, as distinct from bivariate
dependence, is also discussed.
122. Hutchinson, T P 2000. Measuring the success of the Holland
hexagon. Quality and Quantity, 34, 103-110.
J. L. Holland's approach to personality (and careers that are
suitable for different personality types) involves scoring people
on six personality measures and intercorrelating the six scores;
there is a hypothesis about the relative sizes of the fifteen
correlations. Here, some statistics are proposed for describing
how well this hypothesis (and three variants of it) matches an
observed correlation matrix. These statistics are analogous to a
correlation coefficient. A variables-in-common model is given
that justifies the most parsimonious of the hypotheses considered.
123. Hutchinson, T P 2000. ANOVA with skewed data. Environmetrics,
11, 121-124.
Gonzalez and Manly (Environmetrics, 1998, 9, 53-65) discussed
the merits of randomisation tests, in the context of factorial
ANOVA when the data are grossly non-normal. The present note
gives some references to methods that use the ranks of the
observations (including that of Benard and van Elteren),
including computer implementations of these methods. Finally,
the point is made that highly skewed data should not only make
the analyst think seriously about what transformation is
appropriate; it should also lead to consideration of what is the
research question of real concern, and what metric is appropriate
for answering it.
124. Hutchinson, T P 2000. Assessing the health of plants: Simulation
helps us understand observer disagreements. Environmetrics,
11, 305-314.
A dataset on the health of plants, as judged by two raters,
appears to show more disagreement about the relatively healthy
plants than about the less healthy. The bivariate normal
distribution is shown to be a poor description of the data, and a
new bivariate distribution is developed that gives a good fit to
the data. The chief features of the distribution are that it is a
model with latent variables in common (true score plus error for
each rater), and that the scatter of error is greater when the
true score is high than when it is low. As the distribution is
fitted to the data by simulation, an explicit expression for the
joint distribution of the two observed scores is not required.
The software used has ranking and recoding commands of one line
each, so it is easy to ensure the fitted marginal distributions
exactly match the data, and it is unnecessary to estimate
parameters representing the boundaries between the grades of
rating. The method of obtaining the distribution is very
flexible. In one illustration of this, the relation between true
score and the logarithm of the scatter of error is quadratic (not
linear); in another, the true score and the errors have
non-normal distributions.
125. Hutchinson, T P 2000. Measuring the agreement of several
three-letter Holland codes for one person. Journal of Employment
Counseling, 37, 160-162.
A method is proposed for measuring the agreement of several
three-letter Holland codes that have been obtained on one
individual using different methods. The basis of the method is
the scores 4, 2, 1, 0 for Holland themes listed first, listed
second, listed third, and omitted from the three-letter code.
(This is an extension of a proposal by Miller (Journal of
Employment Counseling, 1997, 34, 40-43), who was concerned with
combining several three-letter codes into an average one.)
126. Hutchinson, T P, Chekaluk, E, and Cairns, D 2000. ANOVA applied
to examination scores. Work Study, 49, 104-106.
This paper reexamines data published by Johnnie (Work Study,
1996, 45(6), 22-29) on the performance of two groups of bank
workers on four examination subjects. This leads to a discussion
of the application of the analysis of variance in contexts where
there is one within-person factor (for example, examination
subject), one between-group factor (for example, urban or rural
location of person), and persons constitute a random factor
within the group factor. The analysis of Johnnie's data leads to
a conclusion that differs from the original - that the two groups
differ in mean score on one of the examinations.
127. Hutchinson, T P 2000. A syllabus for transport studies. Road and
Transport Research, 9(2), 62-68. (Brief letter in O.R. Newsletter,
No 364, April 2001, p. 17.)
In reviewing the relative merits of bus and rail systems,
Hensher (Road and Transport Research, 1999, 8(3), 3-21) broadly
favoured the bus. The present writer supports this specific
conclusion, but even more important is Hensher's emphasis on
objective criteria for transport decisions. Suggestions are made
in the present paper for how a degree course in transport could
emphasise rational decision-making, and examples of
technological, operational, political, economic, and
organisational issues are listed.
128. Hutchinson, T P 2000. Graphing the survivorship of bees. Insectes
Sociaux, 47, 292-296.
The statistical distribution of lengths of time (for example,
of the survival of bees) is often of interest. This paper
describes graphical methods that are appropriate for such data,
which typically has a skewed distribution. These methods throw
light on the question of whether hazard rate is constant. Data
published by Visscher and Dukas (Insectes Sociaux, 44, 1997, 1-5)
appears to show increasing hazard rate.
129. Hutchinson, T P 2000. Graphing the death of Escherichia coli.
International Journal of Food Microbiology, 62, 77-81.
The graphical presentation of measurements of the progress over
time of bacterial death or inactivation is discussed.
Specifically, the Weibull distribution may be a useful
generalisation of the exponential distribution. Data from
experiments on the non-thermal death of Escherichia coli reported
by Shadbolt et al. (International Journal of Food Microbiology,
49, 1999, 129-138) are reexamined to support this claim.
130. Hutchinson, T P, and Cairns, D 2000. A note on the syntactic
analysis of Greek translations and compositions. Bulletin of the
International Organization for Septuagint and Cognate Studies,
33, 39-46.
Data from Jobes (BIOSCS, 28, 1995, 19-41) on characteristics
of Greek syntax in seven texts has been reanalysed, using methods
of correlation and multidimensional scaling. It is argued that
(a) the two translations of Esther are very similar to each
other, and (b) the set of papyri are more similar to the four
translations (of Esther and Daniel) than to the two original
compositions (Polybius and Josephus).
131. Hutchinson, T P, and Cairns, D 2000. Discussion of a dataset on the
effect of context on the speechreading of spoken sentences. Journal
of the Academy of Rehabilitative Audiology, 33, 53-61.
Published data from an experiment on the visual perception
(speechreading) of spoken sentences, when received as answers in
question-answer sequences and when received in isolation, are
re-examined. (The data are from Erber, JARA, 25, 1992, 113-122.)
Evidence is found that the extent of the advantage of the
question-answer sequence tends to be less for older subjects. As to
how to measure the difference in performance in two experimental
conditions, arguments are given for preferring the difference in
logits of the success probabilities over either the difference of
probabilities or the ratio of probabilities.
132. Hutchinson, T P 2001. The presumed influence of attention on
accuracy in the Developmental Eye Movement (DEM) test. Optometry
and Vision Science, 78, 700.
Suppose that it is thought that one independent variable
may have an effect at one level of another independent
variable, but not at another level. In this Letter, the point is
made that a researcher may need to carefully consider whether it
is appropriate to test for two "simple" effects, hoping that one
is significant and the other is not, or to test for "interaction".
133. Hutchinson, T P, and Myors, B 2001. Testing Holland's hexagon:
Explanation and criticism. Quality and Quantity, 35, 343-349.
A description is given of Hubert and Arabie's
test of whether an empirical matrix of correlations (between
the six measures of vocational interests commonly abbreviated
R, I, A, S, E, and C) supports Holland's hexagonal (circular
order) model. The description makes clear that the test has severe
limitations. Specifically, the circular ordering RIASEC is only
compared with other circular orderings. Consequently, correlation
matrices arising from a model that is qualitatively different may
be rated very highly by Hubert and Arabie's test; several examples
of this are given.
134. Hutchinson, T P 2001. Correlations may surprise. Teaching
Statistics, 23, 60.
Hignell (Weather, 55, 2000, 150-156) investigated the extent to
which weather interrupts cricket in different parts of Britain. It
is now pointed out that the correlations between the variables in
Hignell's dataset are utterly different from what might be expected.
135. Hutchinson, T P 2001. Discussion of "Flexural fatigue life
distributions and failure probability of steel fibrous concrete" by
S P Singh and S K Kaushik. ACI (American Concrete Institute) Materials
Journal, 98, 419-420.
Singh and Kaushik (ACI Materials Journal, 97, 2000, 658-667)
reported on the fatigue life of steel fibre-reinforced concrete. They
fitted Weibull distributions to several datasets referring to concretes
of different fibre contents being tested at different stress levels.
Singh and Kaushik laid emphasis on the differences between the shape
parameters in different conditions. In contrast, the present discussion
emphasises broad patterns of similarity, making use of the "accelerated
life" hypothesis.
136. Cairns, D, and Hutchinson, T P 2001. Did the gold content of
Cyzicene electrum coins decline over time? A study using
elaboration as a statistical strategy. Revue belge de Numismatique,
147, 51-55.
Was there really a steady decline in the gold content of electrum
coins from Cyzicus, as argued by Smekalova and Djukov (RBN, 145, 1999,
21-35)? We reexamine their dataset, and find that the gold content did
change, but the changes were irregular, rather than being a steady
decline. Further, the average gold content of the lower denomination
coins is higher than for staters.
137. Hutchinson, T P 2001. Partial knowledge and answer-until-correct
tasks in birds and humans. Biometrics, 57, 1251-1252.
Jolliffe and Jolliffe (Biometrics, 53, 1997, 1136-1142) and Ridout
(Biometrics, 55, 1999, 660-662) have proposed models applicable to
memory in coal tits. It is now argued that the birds' task is
analogous to answer-until-correct tests used in educational testing in
humans, and models from the latter context may be used with the birds'
data.
138. Hutchinson, T P 2001. Calculation of the expected lifetime lost
due to an extra risk. Mathematical Population Studies, 9, 209-216.
Methods are given for estimating the average years of life lost
when a person is discovered to be at risk from an extra hazard.
The methods use the probability per year of the extra risk
striking, and the mean and standard deviation of lifetime in the
absence of the risk. The formulae are simple enough that only a
hand-held calculator is needed.
139. Hutchinson, T P, and Gudlaugsdottir, S 2002. Modelling bivariate
ordinal contingency tables arising in studies of interobserver
variation, with application to cervical screening. Journal of
Clinical Epidemiology, to appear.
140. Hutchinson, T P 2002. Should we routinely test for simultaneous
location and scale changes? Ergonomics, to appear.
141. Hutchinson, T P 2002. On decision-making when a risk is
discovered. Journal of Clinical Neuroscience, to appear.
142. Hutchinson, T P, Cairns, D, and Chekaluk, E 2002. The construction
of data to reflect the research objective, and how randomisation
tests make such data usable. Statistical Papers, to appear.
143. Hutchinson, T P, and Gudlaugsdottir, S 2002. Modelling the errors
in scoring the darkness of staining. Analytical and Quantitative
Cytology and Histology, to appear.
144. Hutchinson, T P 2002. The relation between an extreme proportion
and a less extreme proportion, in the context of the comparability
of tests. Journal of Testing and Evaluation, to appear.