Thor NorströmProfessor emeritus

RESEARCH TOPICS

Sociological alcohol research

Sociological suicide research

Evaluations of policy reforms

Methodological issues

CITATIONS ACCORDING TO GOOGLE SCHOLAR(external)

OVERVIEW OF MY RESEARCH

Below is an overview of my research. References to my papers point to the list of downloadable publications, which is found after the research overview. Comments and questions are welcome at totto@sofi.su.se

METHODOLOGY

Modelling time-series data

Much of my research is based on analysing time-series data. A29 addresses various methodological and analytical issues that are central in such analyses. It was written with Ole-Jørgen Skog as a methodological introduction to a series of papers within the European comparative alcohol study – ECAS (A25). All of these papers focus on the link between population drinking and mortality in the EU15 countries during the post-war period (B5 summarizes these papers). We argue that analysis of time-series data is the most feasible approach for assessing the aggregate health consequences of changes in population drinking. We further discuss how aggregate data may also be useful for judging the plausibility of individual-level relationships, particularly those likely to be confounded by selection effects. The aggregation of linear and curvilinear risk curves is treated as well as various methods for dealing with the time-lag problem. A12 is another methodological paper in the same vein.

Lagged response or omitted predictor?

In my analyses of time-series data I have often encountered the time-lag problem, which spurred a certain interest in various approaches to addressing this phenomenon. It is intuitively clear that the omission of a lag structure leads to distorted estimates. In A4 I address the reverse of this problem, one that has received little attention in the literature: suppose that the true model is static (with no lagged response), but that a dynamic Koyck model is specified. This is a quite common approach, implying that the lagged dependent variable is included among the explanatory variables. By means of a simulation experiment, it is shown that the omission of a slow-changing predictor in this situation typically gives rise to the false inference of lagged response, because the residual autocorrelation induced by the omitted predictor is absorbed by the lag parameter. Further, the misspecified dynamics tend to pass unnoticed when applying standard diagnostic tests. Additional tests and an alternative estimation procedure are suggested as remedies. I also model the effects of unemployment and inflation on government popularity to illustrate how the problem may manifest itself in empirical research.

Integrating micro and macro findings

I often try to integrate and compare my macro estimates with the corresponding micro-level findings. However, this is cumbersome as different measures of associations are used at the micro and macro levels. In A10 I present a synthetic approach integrating micro and macro findings. In part 1 I derive the relations between the most common micro and macro measures of associations to make cross-level comparisons possible. In part 2, the approach is applied to the relation between unemployment and suicide.

ASSESSMENTS OF PREVENTION STRATEGIES

High-risk strategy vs. population strategy

The most general paper addressing prevention strategies is A16, which compares the potential of a high-risk strategy with that of a population strategy for preventing acute and chronic alcohol-related mortality. The outcome suggests that a 25% reduction in per capita alcohol consumption (obtained, for example, through increased alcohol taxes) is matched only by a redistribution of drinking of the same magnitude as that obtained through the very strict rationing system that operated in Sweden from 1920 to 1955. A 25% reduction in per capita consumption lies within the limits of recent historical experience, whereas a strict rationing system would hardly be practically or politically feasible.

Cirrhosis mortality was triggered by abolition of alcohol rationing

The crucial importance of per capita consumption for chronic harm (cirrhosis mortality) is documented in A3, based on Swedish time-series data for the period 1931–1980. Much of the analysis in this paper concerns the modelling of the lag structure. Comparing with the experience of other countries, the outcome shows that the rise in Swedish cirrhosis mortality since the mid-1950s was much steeper than should be expected from the increase in per capita consumption. In A5 I address this anomaly, mainly focusing on the impact of the 1955 abolition of the rationing system (which had entailed a maximum purchase limit of 3–4 litres of spirits per month). Analyses of survey data before and after 1955 indicated a marked redistributive impact: the higher the consumption during the rationing, the larger the relative increase after its abolition. Micro-to-macro simulations suggested that most of the excess of the mortality rise was an effect of the redistribution in drinking.

Does the public health perspective apply to tsarist Russia?

Swedish alcohol policy rests on a public health perspective, implying that restricted availability through high alcohol taxes restrains consumption and thereby alcohol-related harm (see A39, with Mats Ramstedt, for a brief overview of Swedish alcohol policies). This perspective has received considerable empirical support from analyses of contemporary data mainly from Europe and North America (see A36 for a review of the link between drinking and harm, and A35 for estimates of alcohol price elasticities based on Swedish data). A more crucial test of the model, though, would be to subject it to data from a context where the cultural and political conditions are not very favourable for its success. In A59 Andrew Stickley and I do find support for this perspective using unique historical time-series data for tsarist Russia spanning the period 1864–1907. Our analyses suggest that the gradual increases in alcohol taxes during this period were associated with a fall in alcohol consumption, which in turn was associated with a fall in alcohol-related mortality. Further, in A52 Tanya Jukkala, Andrew Stickley and I, find a significant association between population drinking and suicide during the same historical period. We obtain a similar alcohol effect estimate when we analyse this association on post-war Russian data. This suggests the continuation of a highly detrimental drinking culture where the heavy episodic drinking of distilled spirits (vodka) is an essential element in the alcohol–suicide association.

ASSESSMENTS OF POLICY INTERVENTIONS

Privatization of alcohol sales – what would happen?

In some papers, research teams in which I participated projected various scenarios implying liberalizations of Swedish alcohol policies, including lowered alcohol taxes (A37) and privatization of retail alcohol sales (A19, A50). These projections, based on a multitude of previous findings, suggest that the negative health consequences of such liberalizations are likely to be significant.

Alcohol sales on Saturdays

A number of studies aim at assessing the impact of more specific policy interventions. A22 reports a significant decrease in traffic crashes after the legal blood alcohol limit for driving in Sweden was lowered from 0.05% to 0.02% in 1990. The findings that Sven Andréasson, Eva Wallin and I report in A31 suggest that a comprehensive programme including responsible server training and stricter enforcement of existing alcohol laws can be effective in decreasing violence in and around bars. The latter also seems to be reduced by reducing bar opening hours. This is suggested by A56, where Ingeborg Rossow and I analyse a series of natural experiments in which bar opening hours in Norway were changed. The evaluation with the strongest design is A32. Here Ole-Jørgen Skog and I assess the impact of the Saturday opening of alcohol retail shops in Sweden by comparing alcohol sales in an experiment area during 17 months with sales in a control area, and find a close to 4% increase in sales. The design is further strengthened in A34, where we included a 13-month post-experimental period when Saturday opening had been implemented in the control area as well. The analyses confirmed the expected equalization in sales between the experiment area and the control area.

MORE DRINKING – MORE DEATHS, ESPECIALLY IN NORTHERN EUROPE

Total (all-cause) mortality is a classic indicator of the overall health of a population. It is therefore of interest to assess its relationship to alcohol consumption, because the net result of alcohol’s supposedly beneficial and known deleterious effects is not obvious. An approach that is of great relevance from a public health perspective is to analyse how the mortality rate responds to changes in overall consumption. I have applied this approach to post-war data for the EU15 countries (A30), Belarus (A46, with Yury Razvodovsky), Russia (A54, B6), Canada (A33) and the USA (A40). By and large, an increase in total consumption is found to be associated with an increase in mortality. However, there is a systematic variation in the strength of this association, so that it is stronger in countries with an intoxication-oriented drinking pattern (e.g. the Nordic countries) than in countries with a more tempered pattern of drinking (e.g. southern Europe). Comparing the estimates with those from corresponding analyses of 19th century data for France (A21) and Sweden (A9) suggests a conspicuous stability across time in this association.

ALCOHOL AND VIOLENCE

The alcohol-violence link is highly contingent on context

One way to avoid confounding is to assess the alcohol-violence link at the aggregate level. A23, based on Swedish post-war time-series data, suggests that violence rates respond significantly to changes in total alcohol consumption. More detailed analyses reveal that the assault rate is related to consumption of beer and spirits in bars and restaurants, while the homicide rate is linked to consumption of spirits in private contexts. The aggregate link between alcohol and violence (homicide) is also addressed from a comparative perspective in a couple of papers (A53; A55, with Jonas Landberg), concluding that the link is stronger in drinking cultures with an intoxication-oriented drinking pattern than in cultures with a more tempered pattern of drinking.

More bars – more violence?

Much of the violence occurs in the vicinity of bars and pubs. Is that because bars attract violence-prone people, or does bar density also have an exogenous effect? Research in this area is typically based on geographical data with which it is hard to disentangle the two effects. However, the hypothesis of an exogenous effect is supported by the findings in A28, based on post-war data for Norway. The findings suggest that an increase in bar density is associated with an increase in violence rates. Bars and pubs thus seem to be suitable targets for interventions to curb violence. Two papers support this notion. A31 (with Sven Andréasson and Eva Wallin) suggests that a comprehensive programme including responsible server training and stricter enforcement of existing alcohol laws can be effective in decreasing bar-related violence. The latter also seems to be reduced by reducing bar opening hours. This is suggested by A56, where Ingeborg Rossow and I analyse a series of natural experiments involving changes in bar opening hours in Norway.

UNRECORDED ALCOHOL CONSUMPTION – TRENDS AND EFFECTS

An indirect method to estimate unrecorded alcohol consumption

Since 2001, unrecorded alcohol consumption in Sweden has been monitored by monthly surveys, but prior to that, and in most other countries, systematic data are scarce. In a series of papers I have elaborated an indirect method to estimate unrecorded alcohol consumption from the discrepancy between the actual level of alcohol-related harms and the harms that are expected on the basis of recorded consumption. I first applied it to Swedish data, uncovering an 80% increase in unrecorded alcohol consumption during the period 1960–1995 (C5). A similar result is obtained for Norway (A24), where comparisons with a series of survey data suggest the indirect estimates correspond well with independent estimates.

Does unrecorded alcohol consumption explain the Russian mortality crisis?

During the period 1990–1994, Russia experienced a mortality crisis unprecedented in peace-time history. The vast literature on this topic offers a host of rival explanations for this, none of which has been generally embraced. For instance, the rise in population drinking, as indicated by the standard alcohol consumption proxy, has been judged as too modest to comprise the major cause. However, when I apply (A54) the indirect method to estimate unrecorded alcohol consumption outlined above, results suggest that actual consumption is grossly underestimated by the commonly used consumption proxy, and that most of the Russian mortality crisis was actually due to the increase in population drinking.

How far do you travel for cheap beer?

In 1995 quotas for private imports of alcohol were markedly increased in Sweden. This encouraged people to take advantage of the lower alcohol prices in Denmark, the main gateway being the ferry line between Helsingør (Denmark) and Helsingborg. Application of the indirect method to estimate unrecorded alcohol consumption (B3) suggests that the ensuing increase in cross-border trading in southern Sweden brought in substitutes for, rather than additions to alcohol obtained from the regular source, Systembolaget. But how far were people prepared to travel to get the cheaper Danish alcohol? Under the premise of substitution (suggested by B3), a drop in Systembolaget’s sales of beer (which was the main beverage imported) can be interpreted as an increase in cross-border trading. The analyses in B4 thus focus on Systembolaget’s county-specific sales of beer over the 24-month period following the change in quotas. The findings suggest that cross-border trading decreased in proportion to the square of the distance to Helsingør (which is consistent with Reilly’s “law of retail gravitation”), and that the maximum travelling distance was about 250 kilometres.

Is unrecorded alcohol extra harmful?

Some studies suggest that consumption of unrecorded alcohol is particularly associated with the risky drinking pattern of binge drinking. In A41, Mats Ramstedt and I thus test the hypothesis that alcohol-related harm rates would respond more strongly to changes in various forms of unrecorded alcohol consumption than to Systembolaget’s alcohol sales. Analyses of time-series data did not lend any support to the hypothesis, although it appeared that a comprehensive consumption indicator (including unrecorded consumption as well as Systembolaget’s sales) was a better predictor of harm than any of the constituent measures.

SUICIDE

Suicide is a classic topic within sociology that I have addressed in some articles. In A17 I elaborate a theoretical link between alcohol abuse and suicide risk within the framework of Durkheim’s suicide theory, a notion that the referees strongly resisted initially. The empirical analyses, focusing on divorce, unemployment and religiosity, in addition to alcohol use, relies on the synthetic approach outlined in A10, that is, a triangulation of micro and macro data modelled with a variety of techniques. The outcome suggests that the alcohol factor is a strong correlate of suicide; religiosity and unemployment have some explanatory power as well, but divorce does not. The aggregate link between alcohol and suicide is also addressed from a comparative perspective in a couple of papers (A7, A18), concluding that the link is stronger in countries with an intoxication-oriented drinking pattern (e.g. Sweden) than in countries with a more tempered pattern of drinking (e.g. France). Further, the possibility of beverage-specific effects is explored in a couple of studies; in A26, Ingeborg Rossow and I conclude that Norwegian and Swedish suicide rates respond to changes in beer and spirits sales (but not wine sales); in A57, Kenji Shibuya, Andrew Stickley and I report similar findings for Japan.

ECONOMIC CHANGE AND LIFE CHANCES IN THE 19TH CENTURY

Some papers assess the impact of economic change (mainly indicated by real wages) on various indicators of life chances in the second half of the 19th century in Sweden. A8 focuses on the impact of various pull and push factors on the emigration flow to the USA. Findings suggest that changes in the Swedish agrarian economy were the primary driving force at the beginning of the period, while the impact of pull factors was not felt until the end of the study period. According to A6, the steady improvement in real wages accounts for the marked reduction in theft criminality. (Further analyses, pertaining to the period a century later, after about 1950, suggest that indicators of economic growth then had the reverse effect on thefts.) Given these findings I found the then-received wisdom in historical demography surprising, namely that after the mid-1800s mortality would have ceased to respond to economic change in the way observed for prior periods. In A9 I thus tested the hypothesis that the beneficial effect of increasing real wages was partially masked by its indirect negative health impact via increased alcohol consumption. It did appear that the wage effect did not attain significance unless alcohol was included in the model.

WELFARE STATE POLICIES AND POPULATION HEALTH

In a couple of articles Tommy Ferrarini, Joakim Palme and I assess the degree to which changes in the generosity of welfare state policies affect population health. The analyses comprise fixed-effects modelling of post-war time-series data for 18 OECD countries. A47 concludes that the more generous the family policies, the lower the infant mortality, while A48 concludes that the more generous the pension rights, the lower the old-age mortality. These findings are summarized and contextualized in A43.

GERONTOLOGY

In a set of papers, Mats Thorslund and I address various topics within gerontology. They are all based on data for 421 persons 75 years of age and older in central Sweden. In A15 we conclude that a commonly used self-assessed global health measure indeed is global as it is significantly associated with each of a large set of more specific health indicators. In A13 we analyse the psychometric properties of various measures of disability. In A14 we analyse the degree to which those measures are related to the use of home help.

Projects