Punched Drunk: Alcohol, Surveillance and the LCBO, 1927-1975

LCBO Surveillance Technologies

Punch Cards, IBM & Statistical Analysis


Data Set and Method for Analysis

The statistical data and table data presented in this work were gathered from all LCBO interdiction files in the RG-36-13 series at the Archives of Ontario for 1929 to 1990. These files contain original applicant letters requesting the LCBO to conduct investigations and the resulting orders, police reports, and other miscellaneous records pertaining to individual cases. The series consists of thirteen boxes, approximately four metres of documents; includes lists of all individuals interdicted through judges’ orders; and a representative sample of the LCBO files on those who had been the subjects of interdiction investigations. The series contains the complete files on over five hundred individual cases investigated between 1953 and 1975. For the purpose of this analysis, we include only data from an individual’s first LCBO interdiction investigation. The reason is to remove the impact that previous records and rulings would have played on the severity of the Board’s action in subsequent investigations. Also, because “all inactive files were destroyed in the mid-fifties,” when the LCBO relocated its offices, we have used only post-1952 files in the analysis to avoid the impact that this removal had on the randomness of the sample. Files were reviewed, ordered by date, and allocated a representative number pertaining to the privacy of those who were the subjects of LCBO investigations. No information is included that would enable anyone listed to be identified, and no subjects were contacted in accordance with the Archives of Ontario’s research agreement (Access Request No. 2004-071). We collected data from files on age, sex, race, geographical location, and involvement of external agencies as reported by the investigator. We also collected data regarding the applicants initiating the LCBO investigations, their relationships to the persons under investigation, the reasons provided by the applicants to justify the investigations, and the job training or occupation reported by the individual under investigation. In four cases applicants provided multiple reasons, which ruled out the selection of a single predominant reason. In those cases we removed the individuals in question from the analysis to maintain the independence of the measured variables.

Data were analyzed data to discover which factors played a statistically significant role in determining the severity of board action. Severity was measured on an eight-point scale: (1) lack of investigation upon an application; (2) investigation with no disciplinary action taken; (3) issuance of a warning letter based on details of the investigation; (4) issuance of a partial order — limiting consumption to either public or private places; (5) issuance of a six-month interdiction order; (6) issuance of a twelve-month interdiction order; (7) issuance of a twenty-four-month interdiction order; and (8) issuance of an indeterminate interdiction order. In all cases, each possibility denotes a uniform increase in the severity of board action, thus allowing for an analysis by means of linear regression. Linear regression presupposes that many factors play partial roles in determining an ultimate outcome. In the case of interdiction, the ultimate outcome is the severity of board action. The model is often simplified into the following equation:

Yi = B0 + B1X1i + B2X2i + … BpXpi + Ei

In this equation, Y represents the outcome variable (the severity of board action), and it is explicable through the predictor variables. For this analysis, the dependent variables were collected from the interdiction files. They include gender, job training or identified occupation of the individual under investigation (occupation), who applied to initiate the investigation (applicant), the reason provided by the applicant, individuals and institutions interviewed or supplying evidence to the investigation (interviews), region of the province the individual lived in, and race. These collected factors were then recoded into the predictor variables (appearing as X1, X2,… Xp), and through the analysis they were given fitted values or parameter estimates (appearing as B0, B1, B2, … Bp) that denote the impact of the given variable; E is the error or model deviation; and i = 1, 2, … n for n observations. The results of the analysis yielded a parameter estimate, or B score, for each tested predictor variable. This score, in significant cases, denoted in the Appendix Figure 1 as p values with *s, shows the extent to which the particular factor impacts the severity of board action on the eight-point scale of possible Board action. For example, a variable showing at least one * and having a B score of –1 denotes that, in cases possessing that variable, the severity of Board action would predictably be one point lower than in similar cases that did not share the specified variable. The analysis also yields an R2 score, found at the bottom of the table beneath each level of the analysis. This score represents, as a percentage, the degree to which the severity of Board action can be explained by the measured predictors within the model. For example, an R2 score of 0.60 would denote that 60 percent of the outcome could be explained by the measured variables, while 40 percent could be attributed to non-measured variables. In this case, the analysis provided a substantial R2 score of 0.715, meaning that the final model of measured variables can explain almost 72 percent of the severity of the board’s response. Significant relationships were found in almost all of the measured factors. The complete results are presented in the accompanying table.