Pre-Trial Detention Under the Young Offenders Act: A Study of Urban Courts
3. Characteristics of Youth Detained by Police: What Factors Affect Police Decisions to Detain? (cont'd)
Multivariate analysis permits conclusions on the effects of each variable in the model while simultaneously controlling for the effects of all other variables. The bivariate analyses in the two preceding sections do not allow us to determine what factors account for police detention at arrest. Our objective is to determine an appropriate combination of predictor or independent variables to help explain the variance in police detention practices.
Logistic regression is appropriate when the dependent, or predicted, variable has two categories. In the following analysis of police detention at arrest (held by police versus released), logistic regression models the natural logarithm of the odds of being in the category of interest as a linear function of the independent variables (details of the regression results are in Tables A.1 to A.8 in the Appendix to this report). The findings are summarized in Table 3.3.
Because of the large variations in findings by court location and (secondarily) in the type of data available, each location has a slightly different model. In addition, the independent variables are operationalized in different ways. "Seriousness or type of current charges" is variously operationalized, including the most serious charge at arrest (see Section 2.4), whether there were indictable offences at arrest or whether the accused was charged with shoplifting or auto theft. Prior record was also operationalized differently depending on the court; in one case it was simply whether the accused had a prior record, in another it was the number of previous convictions, and for other locations, it was a composite variable based on the score generated by a factor analysis of the prior record variables. 
The age of the accused affected police detention decisions in Halifax and Vancouver but in opposite directions. In the former city, younger persons were less likely to be detained whereas in the latter, younger persons were more likely to be held by police, regardless of their other characteristics. The Vancouver finding suggests that child welfare considerations may be a factor in police detention. The race of the young person was associated with police decision-making in Toronto; when other known factors were controlled, black youth were disproportionately detained. Neither age nor race was significant predictors of detention in the total sample. Living with a parent reduced the likelihood of detention in Toronto and Winnipeg and in the sample overall.
In summary, the variables that predicted police detention tended to be legal factors such as the seriousness of the current charges, the number of current charges and the prior record of the young person.
Legal factors associated with the young person had a greater likelihood of influencing police decisions than did demographic or social characteristics of the young person.
The seriousness or type of current charge was most often associated with pre-trial detention, even when all other factors were controlled. This factor was statistically significant in five of the six court locations, Winnipeg being the exception. Even in Vancouver, where 80 percent of the sample were detained, there was a significant relationship between hybrid property charges and police detention decisions: those accused of these less serious property offences were detained in lower proportions than were others. The same variable was also significant in Surrey. In Toronto, the sole type of charge variable that affected detention was whether the current offences involved shoplifting - fewer alleged shoplifters were held compared to other offence types. In Halifax, Edmonton and the total sample, having an indictable offence increased the likelihood of being detained.
In four locations, the number of charges laid at apprehension affected decisions. The Vancouver-area courts showed no relationship between this number and police detention.
Prior record was significant in two courts, even though at the bivariate level the prior record of the accused was significantly related to detention in five courts. In Toronto, the greater the past involvement, the more likely the young person would be detained. In Winnipeg, larger numbers of prior convictions and having a "bad" bail history increased the probability of detention. Similar, but not identical to prior record, is the variable "legal involvement at arrest". In Edmonton those who had no current involvement were detained in significantly lower proportions than were those who had outstanding charges or were on probation. Probationers were detained more often in Surrey than were those with other legal statuses. Thus, indicators of prior convictions were significantly related to police detention in all courts but Halifax-Dartmouth and Vancouver.
As discussed in the last section, information on bench warrants was not routinely captured except in Halifax and Toronto, where (as one would expect) having a warrant significantly contributed to the detention decision. A bail violation among the charges at arrest affected police detention decisions in Edmonton and in the total sample.
In contrast to the bivariate findings in Table 3.2, the number of outstanding charges was significantly related to police detention only in Toronto and the sample as a whole.
All but one model in this analysis was statistically significant at the p<.001 level. In Surrey, the model was significant at the p<.05 level. (See Tables A.1 to A.8 in the Appendix.) The statistic "Nagelkerke r2" in the bottom row of Table 3.3 shows that the amount of variance explained by each model ranged from .16 (16 percent) to .52 (52 percent). The values of the r-squared statistic are more than respectable for Halifax-Dartmouth (.48), Toronto (.52) and Edmonton (.41). The explained variances in Winnipeg and particularly in Surrey are lower than the other courts.
Even though legal factors were thoroughly covered in the regression models, as well as social characteristics of the accused to a lesser extent, they did not explain, or predict, the outcomes in all courts. We must conclude that there are probably other factors affecting police decision-making that were not available to this research.
One possibility is that the operationalization of the independent variables in the regression models was faulty in some way, although this is not likely. A second possibility is that missing data affected our findings.Finally, it is possible that organizational or other environmental characteristics, not capturable in this type of research, contributed to detention decisions. The demeanour of the young person at apprehension, police knowledge of the young person, the area where he or she was apprehended, and the customary practices of police may be among the environmental factors affecting this decision.
 For a more detailed discussion of the construction of the composite prior record variable, see the Appendix.
- Date modified: