No, I’m not working for the organization. I simply asked them to provide me with numbers.
2004: 3.0% / 4.2%
2009: 3.4% / 4.3%
2014: 2.3% / 2.8%
2003: 3.0% / 4.2%
2008: 3.2% / 3.5%
2013: 2.3% / 2.8%
The first category is Assault and the second category is spousal assault. StatsCan analysts helped me with the numbers. I also felt the number 3.4% and 4.3% was wrong for 2009, because police statistics show that both assaults and spousal assaults were declining from 2004 to 2009. Only 61% of respondents responded to the 2009 general social survey on victimization. I’m like there is no way to say that assaults and spousal assaults dropped from 3.4% in 2009 to 2.3% in 2014 and from 4.3% to 2.8% It’s too big of a drop. I feel that if 75% of respondents responded to the survey that it would be my estimates that would be represented.
International Crime Victimization Survey has 3.5% for assaults and threats for 2009. 22% responded to that survey. But I consider the 3.5% for assaults for 2009 to be more in line with reality. They also said 3.0% were assaulted for 2004. The GSS had 4.3% The survey had a 66% response rate for the survey while the GSS had a 75% response rate for that year. I feel that the GSS was more accurate with 80% response rates for the 1988,1993 and 1999 and 2004 surveys, then it started to slip with 75% response rate for 2004, 61% for 2009 and 51.2% for 2014. I was critical of the survey and even told the analysts you have to get the response rate back up to 75% or greater.
Plus I was given this information to clarify
Here is the answer to your questions:
For the first question, we can’t speculate on the reasons, mostly given that we didn’t do any detailed analysis on B&E trends specifically. We did a Juristat specific to B&E a while ago, it’s a bit dated, and it’s using UCR, not GSS, but maybe he could find some answers to his questions:
As for the CI discussion, here is the standard description we now give in our methodology sections:
As with any household survey, there are some data limitations. The results are based on a sample and are therefore subject to sampling errors. Somewhat different results might have been obtained if the entire population had been surveyed.
Confidence intervals should be interpreted as follows: If the survey were repeated many times, then 95% of the time (or 19 times out of 20), the confidence interval would cover the true population value.
So to illustrate this with 2009 and 2014 break and enter (excluding attempts), let’s suppose the true values (the ones we would have got if we had done a census) would be 2.4% in 2009 and 1.9% in 2014.
In 2009, we had 2.66% of Canadians victim of B&E, and the (approximate) confidence interval (CI) was 2.37% to 2.98%, so the supposed 2.4% falls into this bracket (the lower end of it). In 2014, we got 1.73% B&E, with a (approximate) CI ranging from 1.56% to 1.92%. The hypothetic 1.9% also falls within this CI. Of note, the 2009 and 2014 confidence intervals do not overlap, which allows us to say there was a statistically significant difference, so that we’re quite sure that there was an actual true decrease between 2009 and 2014, although the decrease might not be as large as when comparing the exact % we got from the survey.
When we look at the actual numbers we got, it dropped from 2.7% to 1.7%, which indeed seems like a big drop. But in reality, the drop may just as well been from 2.4% to 1.9%, which seems like a more reasonable drop, and our survey data would still be accurate as the true value would have been included in both years CIs. In this example, both the 2.4 and the 1.9 are at the ends of their respective CIs, so this is what I meant here: In 2009, we may have got a sample where the % of victims of B&E was “higher” than in reality, but not necessarily wrong as we always assume the real value would fall within a certain range (CI) from the survey estimate.
Of course, the 2.4% and 1.9% are hypothetical, but it’s just to illustrate that the difference between 2009 and 2014 may be not as large as it looks like (though it could also be as large as it looks like; we don’t know and can’t know for sure), which is why we need to be somewhat cautious when looking at the exact numbers and always keep in mind that the true value will fall within a range from the survey estimate (Of note, in part for this reason, our new standard is now to show confidence intervals in our tables).
As for the victimization by province, this would be a custom data request. we are facing a considerable backlog, since the COVID disruption has channelled almost all of our resources into our core products.