Understanding and Preventing Relapse
Kelly D. Brownell
G. Alan Marlatt
G. Terence Wilson
University of Pennsylvania School of Medicine
University of Washington
University of Oregon and Oregon Research Institute
ABSTRACT." This article examines relapse by integrating
knowledge from the addictive disorders of alcoholism,
smoking, and obesity. Commonalities across these areas
suggest at least three basic stages of behavior change: mo-
tivation and commitment, initial change, and mainte-
nance. A distinction is made between lapse and relapse,
with lapse referring to the process (slips or mistakes) that
may or may not lead to an outcome (relapse). The natural
history of relapse is discussed, as are the consequences of
relapse for patients and the professionals who treat them.
Information on determinants and predictors of relapse is
evaluated, with the emphasis on the interaction of indi-
vidual environmental, and physiological factors. Methods
of preventing relapse are proposed and are targeted to the
three stages of change. Specific research needs in these
areas are discussed.
The problem of relapse remains an important challenge
in the fields dealing with health-related behaviors, par-
ticularly the addictive disorders. This is true for areas of
obesity (Brownell, 1982; Rodin, 198 l; Stunkard & Pen-
ick, 1979; Wilson, 1980), smoking (Lando & McGovern,
1982; Lichtenstein, 1982; Ockene, Hymowitz, Sexton, &
Broste, 1982; Pechacek, 1979; Shiffman, 1982) and al-
coholism (Marlatt, 1983; Miller & Hester, 1980; Nathan,
1983; Nathan & Goldman, 1979).
The purpose of this article is to focus on relapse by
integrating the perspectives of four researchers and eli-
nicians who have worked with one or more of the addictive
disorders (Brownell, 1982; Lichtenstein, 1982; Marlatt,
1983; Wilson, 1980). We will discuss the natural history
of relapse, its determinants and effects, and methods for
prevention. We hope that our collective experience and
different perspectives will aid in developing a model for
evaluating and preventing relapse.
Commonalities and Differences in the
Compelling arguments can be marshaled for both com-
monalities and differences in the addictive disorders.
Many differences exist, both among the disorders and
among persons afflicted with the same disorder. For ex-
ample, genetic contributions to both alcoholism (Mc-
Clearn, 1981; Schuckitt, 1981) and obesity (Stunkard et
al., 1986) suggest separate pathways for their development.
There may be key differences in the pharmacology of
nicotine and alcohol (Ashton & Stepney, 1982; Best,
Wainwright, Mills, & Kirkland, in press; Gilbert, 1979;
Myers, 1978; Pomerleau & Pomerleau, 1984), and food
abuse fits even less neatly with concepts of physical de-
pendency, withdrawal, and tolerance. Treatment goals
also vary, with abstinence the target in some cases and
moderation in others.
Individual differences within the addictions are also
impressive. Variable treatment responses are an example.
There are also striking differences in patterns of use. Some
smokers, alcoholics, and overeaters engage in steady sub-
stance use, whereas others binge. Combinations of phys-
iological, psychological, social, and environmental factors
may addict different people to the same substance. Finally,
different processes may govern the initiation and mainte-
nance of the disorders.
There is also increasing emphasis on commonalities.
One reason is that rates for relapse appear so similar. In
1971, Hunt, Barnett, and Branch found nearly identical
patterns of relapse in alcoholics, heroin addicts, and
smokers. The picture is the same today (Marlatt & Gor-
don, 1985). There may also be common determinants of
relapse (Cummings, Gordon, & Maflatt, 1980). These
factors suggest important commonalities in the addictive
disorders. Progress may be aided by viewing these dis-
orders from multiple perspectives (Levison, Gerstein, &
Maloff, 1983; Maflatt & Gordon, 1985; Miller, 1980; Na-
The notion of commonalities gained support from
expert panels assembled by two government agencies. The
National Institute on Drug Abuse (NIDA) convened a
panel of researchers in alcoholism, obesity, smoking, and
drug abuse and found both conceptual and practical sim-
ilarities in the areas (NIDA, 1979). Similar conclusions
appeared in a more extensive report by the National
Academy of Sciences (Levison et al., 1983). Both reports
noted the importance of relapse and suggested the utility
of combining perspectives from different areas of the ad-
The question of whether the addictions are more
similar than different is difficult to answer. It may be the
case, for example, that there are common psychological
adaptations to different physiological pressures. Nicotine
dependence may be the central issue for a smoker, ex-
cessive fat cells for a dieter, and disordered alcohol me-
tabolism for an alcoholic, but there may be common social
or psychological provocations for relapse, emotional re-
actions to initial slips, and problems in reestablishing
control. Our hope is to expand the information to be
July 1986 ~ American Psychologist
Copyright 1986 by the American Psychological Association, Inc. 0003-066X/86/$00.75
Vol. 41, No. 7, 765-782
focused on relapse by considering both similarities and
differences. In so doing, both conceptual'and practical
ideas may emerge that would not be suggested by the
knowledge available in any one area.
Rates and Definition
Relapse rates for the addictions are assumed to be in the
range of 50% to 90% (Hunt et al., 1971; Hunt & Mata-
razzo, 1973; Marlatt & Gordon, 1980, 1985). This un-
derscores the importance of the problem. However; de-
fining specific rates is difficult. Hidden within these av-
erages is large variability. The rates depend on
characteristics of the addiction, individual variables, the
success of treatment, and so forth.
The figures generally cited for relapse could over-
estimate or underestimate actual rates. Most data are from
clinical programs, so rates are based on those who have
received formal treatment. These figures could overstate
the problem because only difficult cases are seen and be-
cause only one attempt to change is studied (Schachter,
1982). persons attempting to change on their own may
be more successful and may relapse less frequently
(Schachter, 1982). The vast majority of persons who
change do so on their own (Ockene, 1984). These data
could understate the case because clinical programs are
most likely to provide effective treatments. In addition,
#arious criteria are used to define relapse. For example,
relapse in alcohol studies could be defined as days intox-
icated, days hospitalized or jailed, days drinking out of
control, or the use of any alcohol. This points to the need
for standard definitions and for the study of the natural
history of relapse.
Lapse and Relapse--Process Versus Outcome
There are two common definitions of relapse, each re-
fleeting a bias regarding its nature and severity (Marlatt
& Gordon, 1985). Webster's New Collegiate Dictionary
of 1983 gives both definitions. The first is "a recurrence
of symptoms of a disease after a period of improvement."
This refers to an outcome and implies a dichotomous
view because a person is either ill and has symptoms or
is well and does not. The second definition is "the act or
instance of backsliding, worsening, or subsiding." This
focuses on a process and implies something less serious,
perhaps a slip or mistake.
The choice of the process or outcome definition has
important implications for conceptualizing, preventing,
and treating relapse. We suggest that lapse may best de-
scribe a process, behavior, or event (Marlatt & Gordon,
rhis article had its origins in a symposium on relapse at the World
Congress on Behavior Therapy, Washington, I)(2, 1983. This work was
supported in part by Research SCientist Development Award MH00319
from NIMH and by a grant from the MacArthur Foundation to Kelly
D. BrowneU, grant HL29547 to Edward Lichtenstein from NHLBI,
grant AA00259 to G. Terence Wilson from NIAAA, and grant AA05591
to G. Alan Marlatt from NIAAA.
Correspondence concerning this article should be addressed to Kelly
D. Brownell, Department of Psychiatry, University of Pennsylvania, 133
South 36th St., Philadelphia, Pennsylvania 19104.
1985). Webster's defines lapse as "a slight error or slip
?9 . . a temporary fall esp. from a higher to a lower state."
A lapse is a single event, a reemergence of a previous
habit, which may or may not lead to the state of relapse.
When a slip or mistake is defined as a lapse, it implies
that corrective action can be taken, not that control is
lost completely. There is support for this distinction in
smokers (Coppotelli & Orleans, 1985; Mermelstein &
Lichtenstein, 1983) and in dieters (Dubbert & Wilson,
1984). In these cases, different determinants were found
for lapses (slips) and relapses.
The challenge with this approach is defining when
one or more lapses become a relapse. One former smoker
may lose control with the first transgression, whereas an-
other may smoke one cigarette each month and never
lose control. A lapse, therefore, could be defined con-
cretely as use of the substance in the case of smoking and
alcoholism or violation of program guidelines for a dieter.
The individual's response to these lapses determines
whether relapse has occurred. This varies from person to
person and may be best defined by perceived loss of con-
trol. Reliable measures do not yet exist for this assessment.
Research in this area is important for the field.
The Nature and Process of Relapse
Surprisingly little is known about relapse in its natural
state. Most data are from clinical programs where differ-
ent treatments are used with different populations, so it
is difficult to isolate the factors that influence relapse. In
addition, few researchers have done careful evaluations
of patients when they are most likely to relapse, that is,
after treatment has ended. Periodic follow-ups in groups
are the only contacts with patients in most studies, so
repeated, intensive assessments are needed. There would
be great value in learning more about the nature and
process of relapse.
The Need for a Natural History
A metaphor that describes traditional thought on relapse
is of a person existing perilously close to the edge of a
cliff. The slightest disruption can precipitate a fall from
which there is no return'. A person is always on the brink
of relapse, ready to fall at any disturbance. There may
be physiological, psychological, or social causes of the
disturbance, but the outcome is just as final. The first slip
creates momentum so that a complete relapse is certain.
This metaphor may be inadequate.. It does not ex-
plain why a relapse occurs under the same circumstances
that the person managed before. An eating binge may
precipitate relapse in a dieter, but such an individual has
probably recovered from similar binges in the past. A
smoker may relapse after being offered a cigarette, but
there are cases where this same person refused the ciga-
rette or prevented the lapse from becoming a relapse.
Also, the metaphor is based on observations of people
who have relapsed, not those who have not, therefore,
successful recovery is seldom seen.
Information on natural history could address the
question of whether the probability of relapse increases
766 July 1986 ?9 American Psychologist
or decreases with time. If relapse occurs when treatment
"wears off," the probability should increase with time. If
the metaphor used above is valid, the chance of relapse
should increase with time simply because more distur-
bances could occur. One can speculate, however, that a
person learns to cope effectively as time passes and that
those who "survive" beyond the initial period are those
who will succeed. To the extent withdrawal symptoms
precipitate relapse, particularly in smoking and alcohol-
ism, the likelihood of relapse should decrease as the body
adapts to the absence of the addictive substance.
It is in this context that the concept of a "safe" point
arises. This is a point in time before which relapse is
likely and beyond which relapse is unlikely. In the work
of Hunt et al. (1971) on heroin addiction, smoking, and
alcoholism, relapse curves stabilized after the first three
months. It is appealing to conclude that individuals who
abstain for three months are likely to succeed thereafter,
but more recent evidence does not support a specific safe
point (Lichtenstein & Rodrigues, 1977; Wilson & Brow-
nell, 1980). Defining such a point would have important
conceptual and practical implications, so more study on
this topic could pay high dividends. Interpreting relapse
curves may be the first step.
Relapse curves are one type of survival curve. As
such, the figures must be interpreted with several facts in
mind (Elandt-Johnson & Johnson, 1980; Marlatt & Gor-
don, 1985; Sutton, 1979). Group averages do not repre-
sent individuals. Madatt, Goldstein, and Gordon (1984)
found that abstinence rates for smokers after quitting on
the basis of a New Year's resolution were 21% both 4 and
12 months later, implying that relapse rates stabilize and
show a safe point at 4 months. However, different indi-
viduals formed the 21% these two times; some persons
moved from abstinence to relapse whereas equal numbers
moved in the opposite direction. Second, the cumulative
nature of the curves implies that a person who relapses
will remain so; survival curves are negatively accelerating
by their nature. Schachter (1982) noted that cure for many
persons follows several relapses. Third, the probability of
survival for the entire group increases with time because
the persons at highest risk are most likely to leave the
sample. Life table analyses have been designed to deal
with these issues (Elandt-Johnson & Johnson, 1980).
Therefore, it may be possible in future research to develop
a time line for the relapse process and to determine
whether there are "safe" points.
Some information does exist on the natural history
of the addictions. Vaillant's (1983) report on the long-
term progress of 110 alcohol abusers, 71 of whom were
"alcohol dependent," shows the complexity of the issue.
Vaillant's book, and an article by Vaillant and Milofsky
(1982), showed the importance of cultural and ethnic
factors in alcoholism. Many personal and environmental
factors influenced the propensity to drink excessively. It
was clear from these data that a lapse does not necessarily
become a relapse and that this transition has many de-
Schachter (1982) interviewed 161 persons from the
Psychology Department at Columbia University and from
a resort community. In their retrospective accounts, they
reported much higher rates of success at dieting and
smoking cessation than suggested by the literature. Al-
most all successes were achieved without professional aid.
Although Schachter's methods have been questioned (Jef-
fery & Wing, 1983; Prochaska, 1983), he made several
important points. He noted that cure rates are based on
clinical samples and that self-quitters may differ from
therapy-assisted quitters, a notion supported by Di-
Clemente and Prochaska (1982). Second, he found that
many of the successful quitters had made numerous at-
tempts to change before finally succeeding.
Marlatt and Gordon (1980, 1985) have examined
the natural history of the relapse itself. Beginning with a
high-risk situation, their cognitive-behavioral model ad-
dresses the coping process (Figure 1). The absence of a
coping response leads to decreased self-efficacy (Bandura,
1977a, 1977b), then use of the substance, and then the
cognitive phenomenon they label the "abstinence viola-
tion effect." This phenomenon involves the loss of control
that follows violation of self-imposed rules. The end result
of this process is increased probability of relapse. Recent
data from an analysis of relapse episodes in smokers
showed a significant difference in attributions for slips
between subjects who slipped (smoked at least 1 cigarette)
and regained abstinence and those who relapsed (Gold-
stein, Gordon, & Marlatt, 1984). Persons who relapsed
made more internal, characterological attributions for the
slip. This model is useful in conceptualizing the relapse
process from the point at which the person is in a high-
Marlatt and Gordon's (1985) model allows for mul-
tiple determinants of high-risk situations but emphasizes
cognitive processes thereafter. Other factors of a physio-
logical or environmental nature may also be important.
A Cognitive-Behavioral Model of the Relapse Process
Beginning With the Exposure to a High-Risk Situation
Note. Reprinted from Relapse Prevention: Maintenance Strategies in Addictive
Behavior Change (p. 38) by G. A. Marlatt and J. R. Gordon, 1985, New York:
Guilford Press. Copyright 1985 by Guilford Press. Reprinted by permission.
July 1986 ?9 American Psychologist 767
For example, the use of nicotine or alcohol after a period
of abstinence may create a physiological demand for ad-
ditional use. An environmental example is that of a
smoker whose lapse occurs in a social setting where others
are smoking. The resulting cues may provoke further use.
Grunberg and colleagues have found powerful effects of
nicotine on the regulation of body weight and food pref-
erences in both humans and animals (Grunberg, 1982;
Grunberg & Bowen, 1985; G-runberg~ Bowen, Maycock,
& Nespor, 1985; Grunberg, Bowen, & Morse, 1984).
Stopping smoking can create physiological pressure to
change food intake and gain weight. This in turn has
psychological and environmental consequences that can
precipitate relapse. Therefore, it is important to consider
the interaction of individual, environmental, and physi-
ological factors in all stages of the change process.
There is much to be learned about the natural history
of relapse. More descriptive information is needed on
lapses and their associations with relapse. This research
is not easy because the work must be prospective and
because qualitative and quantitative work must be com-
bined. AS an example, Lichtenstein (1984)followed
treated smokers at 1-, 2-, 3-, 6-, and 12-month intervals
with telephone calls. Relapses were preceded by slips for
41 subjects; 19 subjects reported slips but did not relapse.
More information of this nature would be useful.
Stages of Change
Several attempts have been made to divide the change
process into stages (DiClemente & Prochaska, 1982;
Horn, 1976; Marlatt & Gordon, 1985; Prochaska, 1979;
Prochaska & DiClemente, 1982, 1983, 1984; Rosen &
Shipley, 1983). There seems to be a convergence of opin-
ion that at least three fundamental stages exist.
Horn (1976) first proposed four stages of change in
smoking cessation (a) contemplating change, (b) deciding
to change, (c) short-term change, and (d) long-term
change. This is similar to the three-stage models suggested
by DiClemente and Proehaska (I982), Rosen and Shipley
(1983), and Marlatt and Gordon (1985), which involve
the decision and commitment to change, initial change,
and maintenance of change.
Prochaska and DiClemente have done the most
thorough work in this area by evaluating stage models of
smoking cessation and therapy in general. Prochaska
(1979) reviewed 300 therapy outcome studies and pro-
posed five stages, three of which involved "verbal pro-
cesses" and two "behavioral processes." DiClemente and
Prochaska (1982) used this model to compare smokers
who quit on their own to those who used commercial
programs. They proposed the three stages mentioned
above and described six verbal and four behavioral pro-
cesses within the stages. In their recent work, Proehaska
and DiClemente (1983, 1984) suggested five stages: (a)
precontemplation, (b) contemplation, (c) action, (d)
maintenance, and (e) relapse.
More work is needed to test the utility of the various
stage models. They are similar in many respects. Each
has at least one stage where motivation and commitment
are central, followed by initial change and then the
maintenance of change, so we will use these three fun-
damental stages to organize the description of relapse
prevention later in this article. Whichever stage model
prevails, we feel that relapse must be considered in fight
of the stages that precede it. This will draw attention to
the early determinants of relapse and the importance of
the many factors that influence long-term success.
A stage model may also be helpful for relapse itself.
A model might include the time prior to a lapse, the lapse
itself, and the period in which the person does or does
not relapse. The work of Lichtenstein, Antonuccio, and
Rainwater (1977), Cummings et al. (1980), and Shiffman
(1982, 1984) suggests the utility of such an approach.
An important conceptual advance has been the em-
phasis of Prochaska and DiClemente (1982, 1984) on a
circular rather than linear model of change. Linear models
have stages that occur in a specific sequence, with relapse
occurring at the last stage. A circular model shows relapse
leading back to an earlier stage from which an individual
may make another attempt to change. Relapse can be
viewed in a less negative light from this perspective, as
an individual may acquire information or skills that may
be helpful later. This is consistent with Schachter's (1982)
notion that success for most individuals comes after sev-
eral relapses. Taking this to the extreme, one could suggest
that relapse is a necessary step on the path to success.
We do not support this extreme , but we do feel that relapse
may provide valuable experience and that persons who
relapse should be instructed, accordingly.
The Consequences of Relapse
Relapse could provoke a variety of responses in the in-
dividual. It is generally assumed that these responses are
negative, but this may not be true in all cases. This is an
important issue because these responses may determine
the likelihood of success in subsequent attempts to change.
It would appear at first glance that relapse has neg-
ative emotional effects. Disappointment, frustration, and
self-condemnation are apparent in people who relaps e .
Family and friends are unhappy and sometimes angry.
Yet, learning may occur before or during the relapse, so
some benefit may exist. One study tracked depression in
subjects who lost weight and then regained it (Brownell
& Stunkard, 198 I). Depression scores dropped as weight
declined, but returned halfway to baseline as half of the
weight was regained. Although these subjects were not
successful maintainers, the net change in mood was still
There may also be physiological effects of relapse.
When a person stops smoking, the body begins the healing
process, and risk for premature death declines (U.S. De-
partment of Health and Human Services, 1983). Because
there is a dose-response relationship between smoking
and disease, bouts of abstinence may incur some benefit,
so the smoker who relapses may be better off medically
than one who never quit. This is highly speculative, but
it does show that this issue deserves more attention.
The picture may be different in the weight loss area,
768 July 1986 ?9 American Psychologist
where relapse may have detrimental metabolic and health
effects. A recent study found that repeated cycles of weight
loss and regain in animals was associated with increased
metabolic efficiency (Brownell, Greenwood, Shrager, &
Stellar, 1986). As a result, the animals lost weight at half
the rate when they were put on a diet a second time even
though intake was the same on both diets. When allowed
to eat freely, the animals regained at three times the rate
on the second diet than on the first diet. Dieting and
relapse made subsequent dieting more difficult. Epide-
miology studies with humans show positive effects of
weight loss on blood pressure, cholesterol, glucose tol.
erance, and so forth (Simopolous & Van Itallie, 1984).
However, when an equal amount of weight is regained,
the negative effects on blood pressure and cholesterol may
be greater than the positive effects when the weight is lost
(Ashley & Kannel, 1974).
Relapse: Failure or incremental learning? We won-
der whether repeated attempts to change followed by re-
lapse increase or decrease the chance for later success.
There is evidence that persons who have dieted many
times have a poor prognosis (Jeffery et al., 1984; Jeffery,
Snell, & Forster, 1985), although Dubbert and Wilson
(1984) did not find this result. A relapse could be a failure
that strengthens the person's view that the problem is
beyond his or her best efforts. However, relapse may have
positive consequences if the experience somehow prepares
the individual for later success. This more optimistic view
is consistent with Schachter's (1982) suggestion that mul-
tiple attempts occur before many people succeed. A per-
son who relapses may be acquiring information about his
or her weaknesses and may learn ways to prevent lapses
in the future.
This view of incremental learning could be useful
to both professionals and patients. If relapse can be a
constructive experience, experimentation with pro-
grammed relapse might be warranted (Marlatt & Gordon,
1985). This approach involves planning and executing a
relapse that would not occur otherwise, to teach patients
to recover with self-management techniques. This ap-
proach will be discussed in more detail below.
An area that has received little attention is the effect
of patients who relapse on the professionals who treat
them. Following patients through the emotional roller
coaster of success and relapse is discouraging and can
make professionals pessimistic with new patients.
Whether this pessimism is justified depends on perspec-
tive. It is a failure viewed in the short term, but some
long-term effect may occur. Most patients will make other
attempts, and some will succeed.
Determinants and Predictors of Lapse
We make several assumptions here. The first is that there
are similarities in relapse across the addictive disorders
(Marlatt & Gordon, 1985). Our second assumption is
that different processes govern initial change and mainte-
nance (Bandura, 1977a). This assumption has been sub-
stantiated by research on alcoholism (Cronkite & Moos,
1980; Marlatt & Gordon, 1985), smoking (Lichtenstein,
1982; Pomerleau, Adkins, & Pertschuk, 1978; Shiffman,
1982, 1984), and obesity (Brownell, 1982; Dubbert &
Wilson, 1984; Wilson, 1978). The third assumption is
that the risk for relapse is determined by an interaction
of individual, situational, and physiological factors,
The initial attempts to classify relapse situations were
made by Marlatt (1978 ), Marlatt and Gordon (1980), and
Cummings et al. (1980). The Cummings et al. analysis
evaluated 311 initial relapse episodes in drinking, smok-
ing, compulsive gambling, excessive eating, and heroin
addiction. Several determinants emerged, which can be
broadly grouped into individual (intrapersonal) and sit-
uational (environmental) categories. These two categories
are supported by work on smoking (Mermelstein & Lich-
tenstein, 1983; Shiffman, 1982, 1984) and obesity (Dub-
bert & Wilson, 1984). We feel it important to add phys-
iological variables, as their importance is becoming more
clear (Best et al., in press; Brownell, 1982; Lichtenstein,
1982; Myers, 1978; Nathan & Wiens, 1983; Pomerleau
& Pomerleau, 1984).
Individual and lntrapersonai Factors
Negative emotional states. Stress, depression, anxiety,
and other emotional states are related to relapse. Cum-
mings ct al. (1980) found that negative emotional states
accounted for 30% of all relapses. Shiffman (1982, 1984)
evaluated reports of relapse in 264 ex-smokers who called
a telephone hotline service (Stay Quit Line). Subjects were
interviewed soon after the relapse, so reports were recent
even if based only on self-report. Most of the subjects
(71%) had negative affects preceding the relapse, with the
most common mood state being anxiety, followed by an-
ger or frustration, and depression (Shiffman, 1982). Ossip-
Klein, Shapiro, and Stiggens (1984) have also used a tele-
phone hotline to study relapse in smokers. Mermelstein,
Cohen, and Lichtenstein (1983) found that 43% of re-
lapses occur under stress. Pomerleau ct al. (1978) reported
that those who smoke to reduce negative affect are at
increased risk for relapse. A careful study of smokers by
Abrams ct al. (1986) supported these notions by using
physiological, behavioral, and self-report data.
In a study with smokers, Mermelstein and Lichten-
stein (1983) studied both lapses (slips) and relapses.
Lapses were more commonly associated with situational
factors, whereas relapses occurred during negative emo-
tional states or stress events. When the data from these
studies with different addictive behaviors are combined,
it is clear that negative emotional states greatly increase
the chance of relapse. More specifically, negative moods
may increase the chance that a lapse will become a relapse.
Inadequate motivation. It is surprising that so little
work has been done on motivation and commitment. It
would seem that all persons who set out to change are
motivated, particularly those who enter professional pro-
grams. However, there are degrees of motivation, and it
is common for a person to begin the change process in a
burst of enthusiasm without appreciation for the long-
July 1986 ?9 American Psychologist 769