Our analysis was based on telephone interviews with 1,216 respondents derived from the 2005 cycle of the CAMH Monitor, a repeated monthly cross-sectional telephone survey of adults (18 years or older) in Ontario, Canada, conducted by the Centre for Addiction and Mental Health and administered by the Institute for Social Research, York University. This general population survey is part of a long-standing series of annual cross sectional surveys conducted by the Centre of Addiction and Mental Health on a range of topics with foci on tobacco, alcohol, illicit drugs, perscription drugs, gambling and mental health issues. Each year approximately a dozen questions are set aside to obtain public opinion of Ontarians on selected alcohol policy topics. The selection of the alcohol policy questions for each annual survey is based on two main factors: (1) a bi-annual or tri-annual repetition of ‘core’ questions so that monitoring changes in public opinion over time is feasible, and (2) identification of emerging or current issues and selection or development of several questions that are considered most relevant to this topic.

Each monthly cycle of the survey employs a two-stage probability (household, respondent) selection procedure using random-digit-dialing methods and Computer Assisted Telephone Interviewing. To increase the precision of estimates within different areas of the province, the sample was equally allocated among six strata according to area code and the corresponding counties. Within each regional stratum a random sample of telephone numbers was selected with equal probability in the first stage of selection (i.e., households). Within selected households, one respondent age 18 or older was selected according to the most recent birthday of household members. A minimum of 12 call-backs were placed to unanswered numbers and all households who refused to participate on the first contact are re-contacted in order to secure maximum participation. The overall response rate of the 2005 cycle of the survey was 61%. To strengthen the confidence in these data and to ensure that characteristics of the sample were similar to the Ontario population, post-stratification adjustment weights were applied according to the age, sex and regional distribution of the Ontario population (approximately 9,120,000 adults). (See [21] for sampling design details.) The final weights used for the sample were a function of the sampling weight and a post stratification adjustment. Although this procedure does not remove all biases, it does provide a simultaneous adjustment for non-response and non-coverage of households without telephone [22]. Although each yearly cycle consists of 12 independent monthly samples with about 200 completions each, the alcohol policy items were included only in the 6-month period from July to December 2005.

Demographic characteristics of the sample were as follows: 53% were female, the mean age was 45 years (SD = 17.2), 65% were married, 64% had post secondary education (mean years of education is 13.8 years, SD = 2.6), 56% reported an income of $CAD 50,000 or higher, and 76% were past-year drinkers. All cycles of the CAMH Monitor have been approved by the Research Ethics Boards of the Centre for Addiction and Mental Health and York University.

The following 11 alcohol policy opinion items were included in the LCA: (1) alcoholic beverages should have warning labels (yes or no); (2) the government should prohibit wine, liquor and beer advertising on TV (yes or no); (3) the government should prohibit wine, liquor and beer companies from sponsoring sporting or cultural events (yes or no); (4) taxes on alcoholic beverages should be increased, decreased or remain the same; (5) hours of alcohol sales in restaurants, bars or taverns should be increased, decreased or remain the same; (6) efforts to prevent drunken customers at bars, restaurants or taverns to be served should be increased, decreased or remain the same; (7) the legal drinking age should be raised, lowered, remain the same; (8) the number of Liquor Control Board of Ontario (LCBO) stores is too few, too many, or about right; (9) the number of beer stores is too few, too many, or about right; (10) governments should be required to consult with health experts before making legislation or policy changes to the way alcohol is sold (strongly agree, somewhat agree, somewhat disagree, strongly disagree); and (11) the Ontario government should close all LCBO stores, and allow privately-run stores to sell alcohol (strongly agree, somewhat agree, somewhat disagree, strongly disagree). All 11 questions also had a “don’t know” response option.

Questions 1 to 7 focus on counter-promotion and public information (1,2), control of promotion (3), alcohol access (4,5) and interventions (6,7). An underlying rationale of questions 8 to 11 was the tentative initiatives by the Ontario government to increase access to alcohol and possibly privatization of the alcohol retailing distribution system, deliberations that were underway at the time of this survey.

Demographic factors known to be related to alcohol policy opinions [7,12,13] were employed in our analysis. These included sex, age, marital status, years of education, and household income. Alcohol consumption measures included: (1) a drinking status variable with three values: past year drinker, former drinker, and lifetime abstainer, (2) the average number of drinks consumed per week (estimated using the usual quantity by usual frequency approach) and (3) the Alcohol Use Disorders Identification Test (AUDIT), designed to detect problem drinkers at the less severe end of the spectrum of alcohol problems [23,24]. We used the standard cut-off score of 8 or more out of 40 as an indication of problem drinking (range 0−33, mean 3.27).

We used LCA to examine the classes of responding patterns to 11 policy items. Respondents were assigned to classes based on their posterior probabilities for class membership for a particular responding pattern. The 11 policy items were modeled as indicators of a categorical latent variable. A series of unrestricted models (2- to 6-class) were fit to the data using Mplus version 4.2 [25] with maximum likelihood ratio (MLR) estimation. MLR uses random starting values to optimize the parameters. In the initial stage, 500 random sets of starting values were used and a maximum of 2,000 iterations were allowed for each initial stage. In the final stage, 20 optimizations were employed. We began with a 2-class model and proceeded stepwise until the model did not improve further.

To accommodate the complex survey design, the LCA was performed accounting for the sampling weights. To determine the fewest number of classes, several model-fit statistics were examined: the sample size adjusted Bayesian information criterion (ABIC), the Akaike information criteria (AIC), entropy, and the Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (VLMR LRT), which is a test of fit between the model of interest and a model with one less class. Despite the availability of multiple fit statistics, the decision of the most appropriate solution is a function of both statistical and substantive reasoning.

In order to identify socio-demographic characteristics that discriminate among the 5 classes, we employed multinomial logistic regression using independent variables shown to be related to opinions regarding alcohol controls—sex, age, years of education, household income category and AUDIT score. The missing value loss for the multinomial regression due to listwise deletion suggested a negligible impact given that the amount of loss was minimal (91.9% or 1,118 respondents were retained).

To further analyze whether there is a single dimension underlying the LCA results, we conducted a confirmatory factor analysis and constructed a scale based on all items with a factor loading greater than 0.4. Higher values on this scale indicated increased support for alcohol policy control. These analyses were conducted using Stata 9.0, which allowed us to account for the complex survey design.

Table 1 shows results of the 2- to 6-class solutions. Although the VLMR test indicated that the 2-class solution showed the best fit, the BIC values were the lowest (25,260.8) and the entropy values were the highest (0.83) for the 5-class solution. Moreover, the item profiles of the 5-class solution also provided a more interpretable class structure. The mean class probabilities, which indicate the probability of cases being correctly assigned to the class, were 0.944, 0.931, 0.863, 0.890 and 0.930 for class 1 to 5, respectively.

Table 2 shows characteristics of the 5 classes according to the 11 policy items. We have labeled the 5 classes as follows: (1) dedicated liberalizers (class 1; n = 102; 9.5%); (2) moderate liberalizers (class 3; n = 415; 35.9%); (3) moderate controllers (class 4; n = 437; 34.7%); (4) dedicated controllers (class 2; n = 127; 10.3%); and ambivalent (class 5; n = 135; 9.7%).

The dedicated liberalizer and the moderate liberalizer classes represent responding patterns of greater liberalization. Beginning with the “dedicated liberalizer” class, we see that this group, which represents 9.5% of adults, holds strong opposition to alcohol controls: 85.6% believe that the government should not prohibit alcohol advertising on television; 92.1% believe that the government should not prohibit alcohol companies from sponsoring sporting events; and 38.6% believe that taxes on alcohol should be decreased. This class also shows the strongest support for increased access to alcohol: 37.6% indicated that hours of alcohol sales should be increased; 15.6% responded that the legal drinking age should be lowered; 69.0% indicated that there should be more LCBO stores, and 79.9% indicated that there should be more beer stores. This class also shows the highest support for privatizing LCBO stores (33.5%).

The larger “moderate liberalizer” class, representing 35.9% of adults, generally displays a responding pattern similar to the dedicated liberalizer class; however, they are more likely to adopt status-quo positions. Most notably, compared to the dedicated liberalizers, moderate liberalizers are more likely to indicate that taxes should remain the same (65.8% vs. 47.9%); hours of sale should remain the same (82.6% vs. 59.5%); the number of LCBO stores should remain the same (92.9% vs. 27.9%); the number of beer stores should remain the same (96.6% vs. 17.5%). The moderate liberalizer class also departs from the dedicated liberalizers class by showing greater support for raising the legal drinking age (14.7% vs. 6.7%), greater support for increasing efforts to prevent drunken customers from driving (65.5% vs. 46.3%), and lower support for increasing the number of LCBO stores (3.1% vs. 69.0%).

The dedicated controller and moderate controller classes represent responding patterns of greater support for alcohol controls. Beginning with the “dedicated controller” class, which represents 10.3% of adults, we see that his group strongly supports alcohol controls: 98.8% approve of warning labels; 79.4% agree that the government should prohibit alcohol advertisement on television; 72.8% agree that the government should prohibit alcohol companies from sponsoring sporting events; 67.5% indicate that taxes should be increased; 48.7% indicate that hours of sale should be decreased. This class also displays the strongest support for raising the legal drinking age (68.1%), reducing the number of LCBO stores (50.5%) and beer stores (55.4%).

The larger “moderate controller” class, which represents 34.7% of adults, displays a responding pattern similar to the dedicated controllers with the following exceptions. First, they show weaker support for warning labels (80.6% vs. 98.8%), the government prohibition of alcohol advertisements on television (61.2% vs. 79.4%) and sponsorship of sporting events by alcohol companies (40.2% vs. 72.8%). Second, they are more likely than the dedicated controller class to adopt status quo positions: more indicate that taxes should remain the same (71.1% vs. 17.5%); that hours of sales should remain the same (78.4% vs. 23.3%); that the legal drinking age should remain the same (51.2% vs. 15.0%), and that efforts to prevent drunken customers from being served should remain the same (17.4% vs. 3.8%). In addition, this class is more likely than the dedicated controllers to support the status quo of alcohol access. More indicate that the number of LCBO stores should remain the same (96.1% vs. 20.1%) and that the number of beer stores should remain the same (96.3% vs. 21.1%).

The “ambivalent” class, representing 9.7% of adults, captures a responding pattern representative of those with ambivalent, uninformed or ephemeral opinions about alcohol policy. By far, the most distinguishing feature of this class is the high rates of don’t know responses, which range from 8.2% to 65.3% (median = 36.7%). Also, there is some suggestion that they are less knowledgeable about alcohol policy given that 34.4% indicate that they do not know the legal drinking age.

Table 3 shows the demographic and alcohol consumption characteristics of the 5 classes and Table 4 shows results of the multinomial logit model predicting class membership. First, in Table 4, we see that with the exception of years of education, all independent variables (sex, age, income category and AUDIT score) are significantly related to class membership, after adjusting for all other variables. Given that recent research has shown long-term increases in liberalizing attitudes in both Canada and the United States [9,26], we chose the dedicated liberalizers to serve as the reference group in the regression analysis. Thus, each of the four columns in Table 4 provides odds ratios associated with each demographic and alcohol factor for the given latent class relative to the dedicated liberalizers.

As seen in Table 4, compared to the dedicated liberalizer class, those in the remaining four latent classes were less likely to be male and had lower AUDIT scores. Respondents in the ambivalent, moderate and dedicated controller classes were older than those in the dedicated liberalizer class. The odds ratios associated with household income showed that, relative to dedicated liberalizers, respondents in the ambivalent and dedicated controller classes were more likely to reside in households with incomes of less than $30,000. Education was unrelated to latent class membership.

Finally, all respondents were also rank-ordered on the composite unidimensional scale (measuring levels of support for more liberalization in alcohol policy) resulting from the confirmatory factor analysis, which showed sufficiently high goodness of fit values to indicate that the assumption of unidimensionality corresponded well to the data (Comparative Fit Index: 0.93; root mean square error of approximation <0.08; other details not shown). This scale was a composite of all 11 alcohol policy measures, with higher values indicating stronger support for alcohol controls. The five latent classes were also rank-ordered in terms of support of control, with the “Ambivalent” class treated as neutral. A polychoric correlation between the ranked five classes and the scale showed a coefficient of 0.864 (SE = 0.012). This suggests that one single dimension seems to be underlying the five LCA classes.