------- Forwarded Message Follows -------
Date: Thu, 22 Apr 1999 12:47:31 +0200
Reply-to: Kettil Bruun Society <[log in to unmask]>
From: Robin Room <[log in to unmask]>
Subject: Re: Cahalan & Room -- lost classic?
To: [log in to unmask]
Stanton --
Thanks for the kind words.
The book, the typescript for which was completed in 1972, was really the first substantial product of the marriage of two traditions which occurred when Genevieve Knupfer resigned the directorsh
p of the Berkeley group (the California Drinking Practices Study) and Don Cahalan came out from the Washington, DC Social Research Group to head the Berkeley group.
Under Genevieve's direction, the Berkeley group had been a nest of sociologists -- mostly Walt Clark, Ron Roizen and I. (Genevieve had provided cover for us, in the context of the NIMH grants
conomy, by being both a psychiatrist and a PhD in sociology). Our instincts tended to be as "splitters" and nominalists; when Walt wrote up the group's first paper on drinking problems in the gener
l population (QJSA 27:648-668, 1966) the emphasis was on a series of specific alcohol-related problems rather than any overall score. Genevieve's clinical orientation made her a bit more of a "lump
r": she put some emphasis on an "overall problem score", though she was clear that she wanted to talk about "problem drinking" and not "alcoholism" (AJPH 57:973-986, 1967).
Don was a social psychologist, and had just got his PhD (in his 50s, he was a "retread", as he liked to put it) -- which meant that he had spent some time looking at relevant conceptual work i
social psychology. Importantly, Don had also been a journeyperson public opinion researcher for many years, and before that a newspaperman. He thus had great respect for deadlines and getting stu
f out, good or bad, whereas the Berkeley group had had a problem with perfectionism and thus low productivity. Don also appreciated "good stuff" when he saw it, even if it was replacing a draft whi
h he had written. Which made him a good person to have as a boss and coauthor. The basic way of proceeding on Problem Drinking Among American Men was that Don wrote a whole draft, and then I system
tically replaced much of it, section by section.
Don's inclinations when he wrote a book by himself can be seen by looking at Problem Drinkers, his book published in 1970. In my view, the best chapter in that is the conceptual one, attacking
the disease concept and proposing a problem-drinking concept in its place. But for my nominalist sociological taste, this came very close to just replacing one single-lump entity with another, whi
h is why there tend to be multiple dependent variables in PDAAM, or a typological dependent variable.
The same lumping vs. splitting divergence can be seen in the handling of the independent variables. PD, in the social psychological style of the time (e.g., Jessor) lumped items into subscales,
then subscales into scales, and then threw them all in a competitive horse-race -- a grand finale multiple regression. The target was, how much of the variance could you explain? Don did have some
longitudinal data in PD, but was hampered by the limitations in the Time-1 dependent variable, so the grand finale multiple regression was on cross-sectional data. (Don's best extended longitudinal
nalysis was in his dissertation, on pilot study data from Hartford, Conn. While the Berkeley group was kept alive for a long time on the promise of longitudinal analyses, our longitudinal publicati
ns were actually pretty limited before Kaye Fillmore joined the group about 1978). PDAAM, on the other hand, kept different domains of independent variables separate, and used the strategies of run
ing the regression with and without a domain, and of "forcing" variables in, to look at the interactions between the domains in their predictions.
When we wrote PDAAM, we had just begun looking beyond the confines of the US at research traditions elsewhere. At that point, the epidemiological folk at the Addiction Research Foundation were
ery anti-survey, so we had little colleagueship in this respect there. There was already a Nordic tradition, particularly in Finland and Norway (Bruun and Hauge's book on youth surveys in the four
apital cities; Maekelae and others on adult surveys in Finland), and early surveys in Camberwell, England (Griffith Edwards and others), in Sydney, Australia (Margaret Sargent), in the Netherlands (a
ioneering cross-field book called Riskante Gewoonten = Risky Habits), and soon after in Switzerland (Peter Wuetrich). In the U.S., there was Hal Mulford and his colleagues, and a little more periphe
ally Lee Robins and hers (Lee's 1962 paper in Society, Culture, and Drinking Patterns is still very much worth a visit). The early issues of the Drinking and Drug Practices Surveyor, which we star
ed publishing in 1970, give a pretty good picture of what the invisible college looked like.
From a methodological point of view, the part of PDAAM which has stood the test of time best is the use of what would now be called different "models" in the regressions to tease apart relative
ontributions, overlaps in prediction, etc. We must have been relatively early in using this kind of strategy, which is now well recognized. Reflecting public opinion/survey research practice at
he time, we did not routinely report significance tests, but gave general indications about significance in an appendix. This was sensible, but would not pass muster now, and easily leads to overin
erpreting of random results.
What others would call a major methodological deficiency was our lack of formal psychometric underpinnings for our measures, particularly our dependent variables. The "drinking problems scales"
in PDAAM are lineal descendants of scales built by Genevieve Knupfer on the basis essentially of her best judgement. The "severe problems" level got a score of 3, while the "slight problems" level
ot a score of 1, because she thought it should be that way in terms in intrinsic importance and also increased certainty that there was "really" a problem. No-one would have questioned if they had
een scored 2 and 1, but scoring them 3 and 1 made explicit the element of judgement which otherwise would have remained hidden.
When Don came out from Washington, he had a set of problem drinking sclaes already built on the national dataset, using survey research conventions like putting everyone above the mean value on
variable as "high" on some problem. I was influenced enough by Genevieve that this was too nominalist for me, and I persuaded him to throw them out and start again. In building new problem drinki
g measures for him, I did an enormous amount of item analysis on the counter-sorter, and changed the composition of the measures on this basis. Some of this analysis was recorded in the codebooks,
ut in those days doing this was part of the craft of survey analysis rather than something you wrote up and published. As the field began to come under pressure from the new waves in psychiatric ep
demiology, it was left to Mike Hilton years later to do something with the "problem items" a little more psychometrically defensible.
It will be clear from this that I am rather skeptical that the waves from psychiatric epidemiology have moved us forward. A basic problem is the premature lumping of data which is built into th
diagnosis paradigm. Then, too, test-retest reliability is a long way from validity, although the aficionados of this standard tend to obscure this. And no-one ever asks whether it makes sense to m
ke decisions about a score on drinking problems according to its alpha value. In my view, it often doesn't, and PDAAM actually includes some relevant discussion on this, as I remember. What if you
have a model of a dimension for which a and b are alternative, substitute indicators? In that case, the fact that a and b are not highly positively correlated doesn't matter -- in fact, you might p
edict they would be negatively correlated. The different "tangible consequences" of drinking -- family troubles, drinking-driving & police troubles, job troubles -- tend not to cluster very highly
which means that the psychiatric epi. folk look at their "Alcohol Abuse" indicator with much disfavour). But in terms of a particular analysis, they still may belong together conceptually. Down wi
h the hegemony of the correlation coefficient! Up with thinking about the conceptual relations of the variables to be analyzed!
In an international perspective, what PDAAM reflected, under the influence of Genevieve and to some extent Hal Mulford, was the early attention in the US tradition to problems from drinking as w
ll as drinking patterns. Nordic surveys, for instance, have included much less about problems, and mostly did this later. This meant that we were in a position to look at the relation between and
ndividual's drinking patterns and his/her drinking problems, and PDAAM did some of this, although rather awkwardly. Klaus Maekelae picked up on this aspect in his landmark review in the late 1970s on
social consequences of drinking, and he later pioneered the kind of risk-curve approach in this area which came to the fore in Edwards et al., Alcohol Policy and the Public Good (1994).
The part of PDAAM which has been most least picked up in the literature was our various efforts in "contextual analysis", as it was then called (these days something similar is called "hierarchi
al analysis", as in hierarcchical linear modeling). We used three different kinds of measures of the respondent's context in PDAAM: the conventional one of responses about aggregate "social differe
tiations" by the respondent himself (e.g., ethnic identification), census data for the census district of residence, and average scores of those in the respondent's sampling cluster (as with all in-
erson national US surveys, our samples were highly clustered, and we were turning this Achilles' heel into something of use in the analysis). You cite some data from these analyses in your 6th para
raph; in my view, there is much to be learned by going further down these tracks, which to some extent liberate the probability survey from its methodological individualism.
There are two basic problems with PDAAM, of course. One is that it is only about men. It was originally intended simply to "enrich" another sample, including women and the elderly (the sample r
ported in Don's PD), with "positive cases" from the quadrant where most of them were clustered. but there were enogundifferences/advance in measurement that in the end it was more sensible to analy
e the new sample separately.
The other big problem is that it is cross-sectional data (although there is some analysis in it of retrospective data). Which means that one should not be drawing causal conclusions from it: it
can hint, but it cannot prove. Since then, there have been more longitudinal analyses. Even these, however, are shaky ground for drawing the kind of policy conclusions you want to draw from them,
or reasons I don't have time to spell out now. The big step forward in the field in the last 25 years has been the rise of studies of policy change and "natural experiments", which directly measure
the kinds of issues you want to use the PDAAM data as evidence on. They have their limits, particularly concerning long-term effects, but they put us on a much firmer ground to talk about mechanis
s of change that speculating on the basis of data like that in PDAAM.
Thanks for the invitation. I would like to hear reflections by others. Now I've got to go talk to a class about what happened to "social constructivism" in the alcohol field since I wrote a re
iew paper on it 15 years ago! Robin
-----Original Message-----
From: Stanton Peele <[log in to unmask]>
To: [log in to unmask] <[log in to unmask]>
Date: den 21 april 1999 21:14
Subject: Cahalan & Room -- lost classic?
>Dear Robin:
>
>1999 is the 25th anniversary of the publication by the Rutgers Canter of
>Alcohol Studies of "Problem Drinking Among American Men," which I regard
>as the classic work associated with your name. I wonder if you could
>share with the list some reflections on the status of this book -- and
>the alcohol epidemiology field -- on this anniversary.
>
>For me, Cahalan and Room was the first application of modern
>multivariate techniques to large-scale survey research data. The work
>set base rates and identified themes in alcohol epidemiology which it
>has been hard for the field to improve upon.
>
>Among its results were the verification of strong ethnic and
>socioeconomic determinants of drinking behavior and problems -- for
>example, that lower SES respondents were both more likely to abstain and
>to have drinking problems, and that Jews and Ialians had both low
>abstinence and problem rates. Even more determinative of drinking was
>the impact of immediate social groups -- i.e., those with whom the
>drinker associated and drank. While these findings arose from
>traditional sociological areas of concern, C & R firmly established them
>as epidemiologic realities.
>
>A subtheme of C & R was the empirical questioning of the disease model
>of alcoholism, particularly its central focus on loss-of-control
>drinking. C & R did not find loss-of-control to be the wheel horse for
>alcohol problems as Jellinek had indicated. It was Likewise, Cahaln &
>Room did not find progression of drinking problems to be typical.
>
>Indeed, the complex analyses in C & R of the interrelationships and
>ordering of drinking problems identified far more prescient discoveries.
>For example, C & R identified the highest correlating problem with
>"symptomatic drinking" (tolerance/withdrawal) to be psychological
>dependence (with which it was more highly correlated than it was with
>heavy intake). I thought of this relationship particularly when reading
>the recent publications by the WHO/NIJ Cross-Cultural Applicablity (of
>alcohol dependence criteria) Research Project, in which you were a key
>participant, which found that, "Contrary to expectation, descriptions of
>physical dependence criteria appeared to vary across sites as much as
>the more subjective symptoms of psychological dependence" (Schmidt &
>Room, 1999). Not contrary to expectation if you believed Cahalan &
>Room!
>
>C & R also gave empirical grounding to policy concerns such as the
>counterproductive potential of control and regulation of alcohol. For
>example, it found that problems asociated with outburst-type drinking
>were more prevalent in dry areas of the country. This was obviously
>related to issues of availability (if you had to drive to get alcohol,
>you would be more likely to drink quite a bit at once), but it also had
>social-modeling implications. The latter perspective was established by
>findings such as that "both problematic intake and tangible
>consequences. . . have a curvilinear relationship with the father's
>drinking, so that both heavy drinking and, to a lesser extent,
>abstinence on the part of the father predict heavy intake and problems
>on the part of the son" (p. 116).
>
>Robin, I wonder to what extent you feel the field has progressed beyond
>Cahaln & Room. What do you feel are the main new epidemiologic
>discoveries made since 1974? That is, what do we know now that we
>couldn't divine from C & R? More broadly, do you feel that policy is
>better grounded in knowledge than it was in 1974? Are we wiser
>drinkers, and policy makers, based on what was learned from C & R and
>since then?
>
>Thanks, in advance, for sharing your thoughts.
>
>Yes, SP
>
> -----
>To join the KBS-LIST, send the command
> SUBSCRIBE KBS-LIST YOURFIRSTNAME YOURLASTNAME
>To signoff the list, send the command
> SIGNOFF KBS-LIST to [log in to unmask]
>If you experience difficulties signing on or off, write to
>[log in to unmask]
-----
To join the KBS-LIST, send the command
SUBSCRIBE KBS-LIST YOURFIRSTNAME YOURLASTNAME
To signoff the list, send the command
SIGNOFF KBS-LIST to [log in to unmask]
If you experience difficulties signing on or off, write to
[log in to unmask]
|
