Saturday opening of alcohol retail shops in Sweden: an experiment in two phases

Transcription

1 RESEARCH REPORT Blackwell Science, LtdOxford, UKADDAddiction Society for the Study of Addiction 100 Original Article Saturday opening of retail shops in Sweden Thor Norström & Ole-Jørgen Skog Saturday opening of alcohol retail shops in Sweden: an experiment in two phases Thor Norström 1 & Ole-Jørgen Skog 2 Swedish Institute for Social Research, Stockholm University, Stockholm, Sweden 1 and Department of Sociology, University of Oslo, Oslo, Norway 2 Correspondence to: Thor Norström Swedish Institute for Social Research Stockholm University S Stockholm Sweden totto@sofi.su.se Submitted 20 September 2004; initial review completed 13 December 2004; final version accepted 12 January 2005 RESEARCH REPORT ABSTRACT Aim In February 2000, a trial started with Saturday opening of alcohol retail shops in certain parts of Sweden (phase I), and in July 2001, Saturday opening was extended to the whole country (phase II). The aim of this study is to assess the impact of phase II, and to probe previous results regarding phase I. Design Prior to February 2000, all alcohol monopoly outlets were closed on Saturdays. After this date, stores in an experimental area (six counties) were open on Saturdays. In the control area (seven counties) the shops remained closed. To prevent biases due to trade leakage, the experimental and control areas were separated by a buffer area (seven counties). Because continuous evaluations of the trial did not reveal any negative consequences, the Saturday opening was implemented in the whole of Sweden after 17 months. Data and methods The outcome measures included alcohol sales and indicators of assaults and drunk driving. The pre-intervention period covered the time period January 1995-January 2000, phase I of the post-intervention period February 2000 June 2001 (17 months), and phase II July 2001 July 2002 (13 months). The effects of the two phases were estimated through analyses of monthly data (auto-regressive integrated moving-average (ARIMA) modelling) depicting how sales and harm rates evolved in the experimental area compared to the control area during phase I as well as during phase II. Results The analysis uncovered a statistically significant increase in alcohol sales of 3.7% during phase I, and about the same increase during phase II (3.6%). There were no significant changes in any of the assault indicators, neither during phase I nor during phase II. There was a statistically significant increase in drunk driving (12%) during phase I, but no change during phase II. The analyses suggested that the increase during phase I was mainly due to a change in the surveillance strategy of the police. Conclusions The results lend support to the public health perspective in that the increased accessibility to alcohol rendered by Saturday opening also seems to have increased consumption. On the other hand, we could not detect any increase in alcohol-related harm. The question of whether this may be due to insufficient statistical power is discussed, together with some other methodological complications that were highlighted by the study. KEYWORDS Alcohol sales, ARIMA, assaults, drink driving, natural experiment, Sweden. INTRODUCTION Restrictions on hours and days of sales are a widely used alcohol policy instrument aiming at curbing drinking and alcohol-related problems. A large number of studies report that increases in temporal access to alcohol have led to increases of various forms of harm, such as violence (Chikritzhs & Stockwell 2002), traffic accidents (Smith 2005 Society for the Study of Addiction doi: /j x Addiction

2 2 Thor Norström & Ole-Jørgen Skog 1987, 1988), self-destructive behaviour (Northridge, McMurray & Lawson 1986) and drunk driving (Ragnarsdottir et al. 2002). Similarly, decreasing number of days of sales has also been found to have effects. Thus the Saturday closing of Monopoly stores in Sweden 1981 was followed by reduced rates of violence (Olsson & Wikström 1984). Similar findings were reported for Norway (Nordlund et al. 1984). However, there are also studies that report no or ambiguous effects. For instance, McLaughlin & Harrison-Stewart (1992) found no impact on consumption in young males of a temporary increase in trading hours in Western Australia. Ligon & Thyer (1993) found a reduction in drunk driving following a ban on Sunday sales in Athens (Georgia, USA), but this effect was not reversed when the ban was partially lifted (Ligon et al. 1996). A cross-sectional analysis of US states suggested that Sunday sales had an effect on aggregate consumption of sprits, but not of beer (Ornstein & Hannsens 1985). It may be concluded that much of the research on this area suggest an impact of variations in temporal access to alcohol but that the findings are not conclusive. To some extent the inconsistencies may be due to methodological flaws, and the difficulties in conducting controlled evaluations (Stockwell 1995; Babor et al. 2003). Recent policy changes in Sweden of days of sale present an opportunity to apply a more rigorous design for impact analysis. A strict regulation of sales hours has traditionally been a vital part of Swedish alcohol policies. In 1981, a Saturday closing of the alcohol monopoly shops was thus implemented. However, close to 20 years later there was a step in the reverse direction; the Parliament passed a bill stating that the monopoly shops were to be kept open on Saturdays in certain parts of the country during a trial period. If the evaluation of this trial did not reveal any negative effects, Saturday opening should be extended to the whole country, according to the bill. The trial began in February 2000, and the present authors were commissioned to perform the evaluation after about 1 year. The evaluation took the form of an autoregressive integrated moving-average (ARIMA) impact analysis, using time-series data for the designated experimental and control areas. The outcome indicated a slight increase in total alcohol sales (3.3%), but no significant change in various indicators of acute harm, such as assaults and drunk driving [see Norström & Skog (2003) for a more detailed account of the results]. In July 2001, the Saturday opening was thus extended to the whole of Sweden. The extension of Saturday opening can be regarded as an additional experiment, which provides a further basis for probing the results from the evaluation of the trial. By analysing data for the period January 1995 to July 2002, we specifically address the following research questions in this paper (we refer to the trial as phase I, and to the extension of Saturday opening to the whole country as phase II): First and foremost, could the observed effect on alcohol sales during phase I be due to uncontrolled confounders, i.e. factors other than the Saturday opening of shops in the experimental area? Because randomization could not be used in the evaluation, we cannot be sure that there is no systematic difference between the experimental and the control areas. Phase II offers the opportunity to conduct an additional test. If the Saturday effect is real, we should expect that the difference between experimental and control areas more or less disappears after Saturday opening was extended to the whole country. Secondly, is the effect on sales observed in the experimental area representative of the country at large? By comparing the effects of phase I and II, we can address this issue. Thirdly, could the increase in alcohol sales in the experimental area be overestimated due to trade leakage to the buffer area? In the evaluation of phase I, no reduction in sales volume in the buffer area was found, but this does not exclude the possibility that people living in the buffer area added alcoholic beverages bought in the experimental area to their ordinary purchases in their own region. Phase II offers an additional opportunity for testing this possibility. Lastly, could the insignificant results for indicators of acute harm be due to an insufficient number of observations (type II error)? A larger number of observations will reduce the estimation errors and increase the likelihood of uncovering possible effects. DATA AND METHOD The trial comprised three experimental areas, namely Skåne County in Southern Sweden; Stockholm County, which includes the capital and its surrounding districts; and Northern Sweden, including the four northernmost counties in the least densely populated part of Sweden. The control area was located in the middle and southern parts of Sweden, separated from the experimental areas by suitable buffer areas (see Fig. 1). The experimental areas comprised 43% of Sweden s total population; corresponding figures for the control and buffer areas are 34% and 22%, respectively. In addition to alcohol sales, the outcome measures included various alcohol-related harm indicators. More specifically, the following indicators were used: sales of beer, wine and distilled spirits, assaults indoors against women where perpetrator and victim are acquainted (indicator of domestic violence), all assaults indoors, all

3 Saturday opening of retail shops in Sweden 3 Figure 1 Map of experimental areas (black), control areas (crosshatched) and buffer areas (cross-striped) assaults outdoors, all assaults, drunk driving and positive breath analyser tests. Alcohol sales are expressed in litres 100% alcohol per inhabitant 15 years and above per month (Data provided by Systembolaget, which has a monopoly on retail sales of alcohol). The data on assaults and drunk driving refer to offences reported to the police per population 15 years and above (Data provided by Swedish National Council for Crime Prevention). The supplementary indicator of drunk driving is the number of breath analyser tests exceeding the legal blood alcohol limit, i.e. 0.02% (Data provided by National Laboratory of Forensic Science). Any impact of the Saturday opening on drinking and harm is likely to be concentrated to the weekend days. Therefore, the harm indicators (offences) refer to number of cases per month that have occurred on Saturdays or Sundays. The breath analyser data have an even more differentiated time window (days and hours). However, the analyses of sales data are based on regular monthly data. The pre-intervention period covers the time period January 1995 to January The trial was run during the period February 2000 to June 2001, and in July 2001 the Saturday opening was extended to the whole country. The end-point for most of our data is July 2002, inclusive, implying a post-intervention period of 17 months for phase I of the Saturday opening (the trial) and 13 months for phase II (the extension). For breath analyser tests the end-point is June As in the previously reported evaluation of phase I (Norström & Skog 2003), we evaluated the effects in two steps. First, we use a graphical procedure, based on the correlation between experimental and control area prior to February Secondly, we use parametric models with intervention dummies and moving average and autoregressive parameters for the noise term in order to obtain numerical estimates of the Saturday effects in phases I and II. The graphical procedure was as follows. Each outcome measure (sales, assault rates and drunk driving) for the experimental area was regressed on the corresponding measure for the control area, using an ARIMA model for the time period January 1995 January (Clearly, no intervention dummies were included in this analysis.) Based on this model, the outcome measure for the experimental area could be predicted on the basis of the corresponding measure for the control area. The difference between the actual measure and the predicted measure could then be calculated for the entire period, i.e. both prior to and after February The plot of this difference between actual and expected outcome measures gives a visual representation of, first, the effects of phase I and, secondly, the extent to which phase II closed the gap (if any) between experimental and control area. If the increase in alcohol sales in the experimental area during phase I was due in fact to causes other than the Saturday opening (i.e. caused by confounding factors), we would expect the same confounding factors to be present during phase II as well. In that case, the gap between observed and predicted sales should prevail during phase II, as in Fig. 2a. On the other hand, if the observed increase during phase I is a true effect of the Saturday opening, and the effect is of the same order of magnitude in the control area during phase II, we should expect a pattern similar to Fig. 2b. However, the size of the effect could be different in the experimental and control area. For instance, if the increase in the experimental area during phase I was due to some extent to purchases made by people living in the buffer area, the observed effect could be larger in the experimental area than in the control area during phase II. An example of the latter pattern is given in Fig. 2c, while Fig. 2d illustrates a case where the effect is largest in the control area. The latter pattern could obtain if economic or cultural differences between the regions produced larger effects in the control area.

4 4 Thor Norström & Ole-Jørgen Skog a. b. Difference between experimental- and control area Difference between experimental- and control area c. d. Difference between experimental- and control area Difference between experimental- and control area Figure 2 Four hypothetical outcomes. (a) The increase observed in the experimental region is due to confounding. (b) The Saturday effect is the same in both experimental and control regions. (c) The Saturday effect is larger in the experimental region than in control region. (d) The Saturday effect is smaller in the experimental region than in control region Besides the graphical method, we also used parametric models in order to obtain numerical estimates of the intervention effects. In the evaluation of phase I, the Saturday effect was estimated according to the following model: lne t = a + b 1 lnc t + b 2 I t + cd t + N t where E and C denote sales in experimental and control area, respectively. I is an intervention dummy coded zero prior to February 2000, and one from this date onwards. D represents a dummy variable for differences in seasonal patterns (in this case December), and N is the noise term. The parameter b 2 can be interpreted as a measure of the relative increase in sales volume in the experimental area during phase I due to Saturday opening of shops. The log transformation was required due to heteroscedasticity in the raw data. When all shops were opened on Saturdays in July 2001 (phase II) a similar, but not necessarily equally large effect could be expected in the control area. To estimate this effect the original model was extended by including another intervention dummy, J, coded zero prior to July 2001 and one thereafter. The model then becomes: lne t = a + b 1 lnc t + b 2 I t - b 3 J t + c D t + N t The parameter b 2 still measures the effect of the Saturday opening in the experimental area (phase I), while b 3 can be interpreted as a measure of the effect of phase II. (Note that if the gap between the experimental and control areas are reduced in phase II, the sign for the phase II effect will be negative.) In order to test whether the effect is the same in experimental and control areas, we need to test the nullhypothesis H 0 : b 2 b 3 = 0. This was obtained as follows. The dummy variable P is defined as: P t = I t - J t

5 Saturday opening of retail shops in Sweden 5 and P is consequently equal to 1 during the period February 2000 to July 2001, and zero before and after these dates. From the original equation we then obtain the following: lne t = a + b 1 lnc t + b 2 P t + d J t + cd t + N t where, from the definition of P, the parameter d can be seen to be equal to: d = b 2 - b 3 We can evaluate the above hypothesis by testing whether the parameter d is different from zero. These models were used for all outcome measures to obtain estimates of the Saturday effects in the experimental and control area, and to test the difference between the two effects. The parameter estimates b 2 and b 3 can be converted to estimates of increase in percent by taking antilogarithms, subtracting 1, and multiplying the result by 100. The structure of the noise term was estimated by including moving average and autoregressive terms, both ordinary terms (AR and MA) and seasonal terms (SAR and SMA); these are reported in the tables. The fits of the models were evaluated using the Box Ljung portmanteau test of the first 12 lags of the residuals, Q(12), which are reported in the tables, and by graphical inspection of the residuals (not reported). The models outlined can be estimated on the raw data, on seasonally differentiated data (lag 12 months), as well as on regularly differenced data (lag 1 month). It turned out that seasonal differencing gave the best fit for the sales data, while the models could be estimated from the raw data for assault and drunk driving. However, the results obtained with the alternative approaches (not reported) were not very different from the results presented here. For a more detailed account of the technique for ARIMA modelling, see Box & Jenkins (1976). In the previously published analysis, we tried to estimate possible buffer trade effects between the experimental and buffer area by looking for possible reductions in sales volume in the buffer area after February No significant changes were found. However, people in the buffer area could increase their purchases in the experimental area without decreasing their purchases in their own region. We have tried to evaluate this possibility by analysing the changes in sales volume in the buffer area after July If inhabitants in the buffer area behaved in the way just suggested, one could expect that their purchases in the experimental area would decrease and that their purchases in their own area would increase more than the increase in the control area after July In order to investigate this possibility, the sales volume in the buffer area (B) was regressed on sales volume in the control area, on a dummy coded one during the trial period from February 2000 to June 2001 and zero otherwise (P), and on another dummy coded one from July 2001 and onwards, and zero before that (J). The resulting equation is: lnb t = a + b 1 lnc t + b 2 P t + b 3 J t + cd t + N t The first dummy should pick up the decrease during the trial period (if any), while the latter should pick up any excess increase after July 2001 in the buffer area, compared to the control area. RESULTS Alcohol sales Border trade from Norway represents a potential methodological complication for the estimation of the effects of phase II. The control area included six shops located close to the Norwegian border, and the experimental area included one such shop. A substantial fraction of the alcohol sales in these shops goes to Norwegians. To check whether the extension of the Saturday opening magnified the border trade to Norway, the sales in these six shops were compared to the sales in the remaining shops in the control area. As can be observed (Fig. 3), the sales increase following the opening on Saturdays in July 2001 was substantially higher in these six shops. For the period July 2001 to July 2002, the increase was 61% higher than in the rest of the control area. Although the sales volume in the six shops amounts to only about 5% of the overall sales volume in the control area, the Norwegian border trade clearly needs to be controlled in the analysis, in order to prevent biases in the results. The solution chosen was to exclude the shops in question from the analysis six were excluded from the control area, and one from the experimental area. Index Figure 3 Ratio of alcohol sales in six shops close to the Norwegian border to alcohol sales in the remaining shops in the control region. Index with June 2001 = 100

6 6 Thor Norström & Ole-Jørgen Skog a. Beer.02 c. Spirits b. Wine d. Total alcohol Figure 4 Difference between observed and predicted alcohol sales in the experimental area. Phases I and II of Saturday opening indicated by vertical lines Figure 4 presents the difference between observed and predicted alcohol sales for the experimental area during the period January 1998 to July For beer, wine and distilled spirits there was a marked increase during the trial period. During phase II, the discrepancy between the experimental and control area became much smaller, suggesting that Saturday opening had a substantial effect on sales volume in the control area as well. The numerical estimates are reported in Table 1. The estimated effects of phase I correspond fairly well with those obtained in the previous report (Norström & Skog 2003). The estimated rate of increase is significant for all beverage types. For beer the parameter estimate (0.073) corresponds to 7.6%; for wine (0.025), to 2.5%; for spirits (0.036), to 3.7%; and for total alcohol (0.036), to 3.7%. The estimates for phase II are somewhat smaller for beer and spirits. The difference is significant at the 5% level. For wine, the estimate for phase II is slightly higher, but the difference is not significant. For total alcohol the phase II estimate is practically the same, and the difference is not significant. These beverage-specific differences between the effects of phase I and II might suggest that the effect was actually slightly larger in the experimental area than in the rest of the country, or they could be due to an overestimation of the phase I effect owing to purchases made in the experimental area by people living in the buffer area. Because the differences are observed for beer and spirits only, the latter explanation would imply that we should expect changes in the buffer area during phase II specifically for these beverage types. Table 2 sheds some light on this issue, reporting results of the analysis of sales in the buffer area. First we observe that no significant decrease could be observed during phase I. During phase II, however, there was a small but significant increase in the sales of beer (1.8%) and spirits (1.2%). For wine there was a small decrease, but this effect is insignificant. The increase in total alcohol was small and

7 Saturday opening of retail shops in Sweden 7 Table 1 Estimated effects of the Saturday opening on beer, wine, spirits and total alcohol sales during phases I and II. Double logarithmic ARIMA models estimated on seasonally differenced data for the period January 1995 July Beer Wine Spirits Total alcohol b SE (b) B SE (b) b SE (b) B SE (b) Saturday phase I 0.073*** *** *** *** 5 Saturday phase II 0.052*** *** *** *** 5 Difference I II 0.021* -8 (NS) 0.014* 1 (NS) Control area Constant < SMA(1) Q(12) *P < 0.05; ***P < 1. NS: not significant (P > 0.05). Q(12): Box Ljung test for autocorrelation, 12 lags. None are significant at the 1% level. Table 2 Estimated effects of the Saturday opening on beer, wine, spirits, and total alcohol sales in the buffer area during phase I and II. Double logarithmic ARIMA models estimated on seasonally differenced data for the period January 1995 July Beer Wine Spirits Total alcohol b SE (b) B SE (b) b SE (b) b SE (b) Saturday phase I -1 (NS) 5-9 (NS) 6-6 (NS) 3-6 (NS) 4 Saturday phase II 0.018* (NS) * 4 4 (NS) 6 Control area Constant AR(1) SMA(1) Q(12) *P < 0.05 NS: not significant (P > 0.05). Q(12): Box Ljung test for autocorrelation, 12 lags. None are significant at the 1% level. insignificant. Hence, the pattern corresponds to what one should expect if the effect was slightly larger in the buffer area, as was suggested earlier. However, it should be stressed that other factors could also explain the observed result. Harm indicators In this section we address the issue of whether the increased alcohol sales spurred by the Saturday opening also led to increased rates of alcohol-related harm. Turning first to the assault indicators, the difference between observed and predicted values for all assaults is displayed in Fig. 5a. No effect can be observed, neither of phase I nor of phase II. (The outcome is by and large the same for the remaining assault indicators; these graphs are therefore not shown.) The outcome of the model estimations confirms the graphical evidence. The parameter estimates are reproduced in Table 3, and none of the effects are significant. Hence, the Saturday opening and the increase in consumption did not have any noticeable effect on assault rates. In the case of drunk driving offences, we found a significant increase in the experimental area in the previously reported evaluation of phase I. However, in that context we hypothesized that this could have been an artefact. In particular, there was intensified screening for drunk driving in the vicinity of alcohol monopoly shops at Saturday opening hours during the trial in a large part of the experimental area (Stockholm). Because one can expect a concentration of drunk driving in such areas due to drinking the night before, an increase in such offences might well be due to this local change in the surveillance strategy of the police. If this is so, we should expect the gap between experimental and control area to prevail during phase II for drunk driving offences (unless the police in other regions tend to follow the same strategy). As can be seen from Fig. 5b, this appears to be the case. The numerical estimates, reproduced in Table 4, confirm this impression. There was an increase during phase I (0.113), corresponding to 12.0%, but practically no effect of phase II. The difference between the two phases is significant at the 1% level. The outcome for positive breath analyser tests during the period Saturdays 10

8 8 Thor Norström & Ole-Jørgen Skog a. Assaults, total 6.00 d. Positive breath analyser tests. Saturdays 10 AM - Saturdays 2 PM b. Drunk driving offences c. Positive breath analyser tests. Saturdays 10 AM - Sundays 2 PM e. Positive breath analyser tests. Saturdays 2 PM - Sundays 2 PM Figure 5 Difference between observed and predicted harm rates in the experimental area. Phases I and II of Saturday opening indicated by vertical lines a.m. Sundays 2 p.m. is in the same direction (Fig. 5c and Table 4). As can be seen further from Fig. 5d e, the increase in positive breath analyser tests is in fact confined to the Saturday opening hours of the alcohol monopoly shops (10 a.m.-2 p.m.), an increase that persists in phase II, while there is no noticeable increase after 2 p.m. on Saturdays, neither in phase I nor phase II (Table 4). This strengthens the hypothesis that the effect observed for drunk drinking is an artefact caused by a local change in surveillance

9 Saturday opening of retail shops in Sweden 9 Table 3 Estimated effects of the Saturday opening on assault rates on Saturdays and Sundays (female victims indoor, all victims indoor, all victims outdoor, and all assaults) during phase I and II. Double logarithmic ARIMA models estimated on raw data for the period January 1995 July Females indoor All indoor All outdoor Total b SE (b) B SE (b) b SE (b) b SE (b) Saturday phase I 6 (NS) (NS) (NS) (NS) Saturday phase II (NS) (NS) (NS) (NS) Difference I II (NS) (NS) (NS) (NS) Control area Constant SAR(1) Q(12) NS: not significant (P > 0.05). Q(12): Box Ljung test for autocorrelation, 12 lags. None are significant at the 1% level. Table 4 Estimated effects of the Saturday opening on drunk driving on Saturdays and Sundays, and on positive breath analyser tests on Saturday after 10 a.m. and 2 p.m., respectively, during phases 1 and 2. Double logarithmic ARIMA models estimated on raw data for the period January 1997 July 2002 (for breath analyser to June 2002). Drunk driving Saturday Sunday Breath analyser Saturday 10 a.m. Sunday 2 p.m. Breath analyser Saturday 2 p.m Sunday 2 p.m. b SE (b) b SE (b) b SE (b) Saturday phase I 0.113** 0.036*** (NS) Saturday phase II (NS) (NS) Difference I II 0.131** 0.239*** (NS) Control area December dummy Constant SAR(1) Q(12) **P < 0.01; ***P < 1; NS: not significant (P > 0.05). Q(12): Box Ljung test for autocorrelation, 12 lags. None are significant at the 1% level. strategy, rather than a real increase due to increased consumption. DISCUSSION The extension of the Saturday opening of the alcohol monopoly shops from an experimental area to the whole of Sweden provided a possibility to probe the results from the evaluation of the trial. By and large, these results were corroborated; the extension was thus followed by an increase in alcohol sales of about the same magnitude as was the case for the trial, close to 4%. The non-existent effects on the harm indicators were also confirmed by the present analyses. Although the experimental design used must be regarded as methodologically strong, it does have some potential weaknesses that were highlighted in the present application. It thus appeared that the intervention had some unanticipated effects that, if gone unnoticed, would have biased the outcome. One such effect was the increased border trade to Norway in the control area that was spurred by the Saturday opening; another distorting effect was the change in the strategy of police surveillance of drunk driving offences. Another critical issue concerns the risk of type II errors. The lack of any intervention effect on most harm indicators may be due to insufficient statistical power. For example, it takes at least a 5% increase in assaults if the effect is to become statistically significant, granted the SE of 2.22% of the estimated intervention effect (Table 3). This is probably more than should be expected from a 4% increase in alcohol sales. The elasticity between sales and assaults in Sweden has previously been estimated at approximately 0.5% (Norström 1998). Hence, the observed increase in alcohol sales could at

10 10 Thor Norström & Ole-Jørgen Skog most be expected to bring about an increase of approximately 2%. The reform evaluated in this study had its main effect on off-premise consumption. Some studies have suggested that alcohol-related violence is more strongly connected to on-premise consumption than to off-premise consumption (Lenke 1990; Norström 1998). This might suggest that the effects on assaults of the present reform could be even smaller than 2%. Furthermore, the reform may have had some indirect effects on on-premise consumption, for instance by reducing on-premise sales when it was no longer necessary to visit bars and pubs in order to obtain alcohol on Saturdays. The net effects of a substantial increase in off-premise consumption and a small decrease in on-premise consumption on assaults could be close to zero. Whether or not a similar mechanism might apply to drunk driving is an open question. Due probably to differences in blood alcohol concentration (BAC) limits and law enforcement the prevalence of drunk driving is very different in Scandinavia and the United States (Berger et al. 1990), and the connection between drunk driving and drinking context (private versus public) may also be different. On the more substantive side, we can conclude that the results support at least one of the central tenets of the public health perspective (Edwards et al. 1994), that increased availability of alcohol this time in the form of Saturday opening tends to spur consumption. References Babor, T. F., Caetano, R., Casswell, S., Edwards, G., Giesbrecht, N., Graham, K., Grube, J. W., Gruenewald, P. J., Hill, L., Holder, H., Homel, R., Österberg, E., Rehm, J., Room, R. & Rossow, I. (2003) Alcohol: No Ordinary Commodity. Oxford: Oxford University Press. Berger, D. E., Snortum, J. R., Homel, R. J., Hauge, R. & Loxley, W. (1990) Deterrence and prevention of alcohol-impaired driving in Australia, the United States and Norway. Justice Quarterly, 7, Box, G. E. P. & Jenkins, G. M. (1976) Time Series Analysis: Forecasting and Control. Revised Edition. San Francisco: Holden-Day. Chikritzhs, T. & Stockwell, T. R. (2002) The impact of later trading hours for Australian public houses (hotels) on levels of violence. Journal of Studies on Alcohol, 63, Edwards, G., Anderson, P., Babor, T., Casswell, S., Ferrence, R., Giesbrecht, N., Godfrey, C., Holder, H., Lemmens, P., Mäkelä, K., Midanik, L., Norström, T., Österberg, E., Romelsjö, A., Room, R., Simpura, J. & Skog, O.-J. (1994) Alcohol Policy and the Public Good. Oxford: Oxford University Press. Lenke, L. (1990) Alcohol and Criminal Violence Time Series Analyses in a Comparative Perspective. Stockholm: Almqvist & Wiksell. Ligon, J. & Thyer, B. A. (1993) The effects of a Sunday liquor sales ban on DUI arrests. Journal of Alcohol and Drug Education, 38, Ligon, J., Thyer, B. A. & Lund, R. (1996) Drinking, eating, and driving: evaluating the effects of partially removing a Sunday liquor sales ban. Journal of Alcohol and Drug Education, 42, McLaughlin, K. L. & Harrison-Stewart, A. J. (1992) The effect of a temporary period of relaxed licensing laws on the alcohol consumption of young male drinkers. International Journal of Addiction, 27, Nordlund, S., Hauge, R., Nordlie, O., Ihlen, B. M., Irgens-Jensen, O. & Horverak, Ø. (1984) Virkninger av lørdagsstengte vinmonopolutsalg. En Samlerapport. [Effects of Saturday closing of liquor retail stores: a summary report, in Norwegian.] Oslo: National Institute for Alcohol Research. Norström, T. (1998) Effects on criminal violence of different beverage types and private and public drinking. Addiction, 93, Norström, T. & Skog, O.-J. (2003) Saturday opening of alcohol retail shops in Sweden: an impact analysis. Journal of Studies on Alcohol, 64, Northridge, D. B., McMurray, J. & Lawson, A. A. (1986) Association between liberalisation of Scotland s liquor licensing laws and admissions for self poisoning in West Fife. BMJ, 293, Olsson, O. & Wikström, P. O. H. (1984) Effects of the experimental Saturday closing of liquor retail stores in Sweden. Contemporay Drug Problems, 11, Ornstein, S. I. & Hannsens, D. M. (1985) Alcohol control laws and the consumption of distilled spirits and beer. Journal of Consumer Research, 12, Ragnarsdottir, T., Kjartansdottir, A. & Davidsdottir, S. (2002) Effect of extended alcohol serving hours in Reykjavik, Iceland. In: Room, R., ed. The Effects of Nordic Alcohol Policies, pp NAD Publication 42. Helsinki: Nordic Council for Alcohol and Drug Research. Smith, D. I. (1987) Effect on traffic accidents of introducing Sunday hotel sales in New South Wales. Australia Contemporary Drug Problems, 14, Smith, D. I. (1988) Effect on traffic accidents on introducing Sunday alcohol sales in Brisbane, Australia. International Journal of Addiction, 23, Stockwell, T. R. (1995) Do controls on the availability of alcohol reduce alcohol problems? In: Stockwell, T. R., ed. An Examination of the Appropriateness and Efficacy of Liquor Licensing Laws across Australia., pp Canberra: Commonwealth Department Human Services and Health, Australian Government Publishing Service.