Possum has a very interesting argument and analysis that suggests ‘no’. Well, at least, ‘probably not’.
What he does is compare the rate at which houses with insulation burn down and tries to differentiate between the stock and flow effects that we might observe. He actually comes up with the counter-intuitive result that the Garrett program has reduced the risks of house fire.
Under Scenario 1 where 90% of fires are attributable to new installations, 1 in 909 installs lead to fire. Under Scenario 2 it’s 1 in 1636 installs lead to fire while under Scenario 3 it’s 1 in 8182 installs.
Under the Garrett insulation program, the rate is 1 in 11,828 – a much smaller rate of fires than what existed before the program.
The Insulation Program Safety Multiple is simply the Garrett program rate divided by the 2008 rate – it shows how many times safer the Garrett program is compared to each of the three scenarios for 2008.
While I’m not sure that his final analysis is correct, I do think he is on the right track and asking exactly the right question; what is the marginal increase in housefires due to the Garrett insulation program?
First we have to think about the ‘stock effect’. Each year there will x number of housefires due to existing insulation. Second there is a flow effect. Each year there will be y number of housefires due to newly installed insulation. We will probably only observe z housefires due to insulation in any one year, where z = x + y. Possum cites the federal government as saying that each year there are 80 – 85 housefires associated with insulation. So 82.5 = x + y (to take an average).
Possum tells us that in 2008 there were 3,183,265 houses with insulation and the rate of increase each year was 67,500. Okay. The Garrett program insulated 1,100,000 houses since July 2009. (That seems a bit high.) There are 93 housefires associated with that program – maybe more, but lets stick to Possum’s numbers. Remember also that this program only started eight months ago – so those 93 house fires are not an annual number. What we don’t know is whether that 93 includes the stock of house fires that we would normally experience or not. But going back to our equation we have x + 93 = z so far for the year. Adding to the complication is that some houses with insulation already were reinsulated. We don’t know whether that reinsulation has added to the risk of housefire or not.
Possum works out rates – what is the rate of increase in housefire risk due to insulation? The analysis here turns on how we allocate housefires to either the stock or the flow. The increase in housefires does look very large, but the flow in the past eight months has been massive relative to previous years (1,100,000 v 67,500). Possum is able to show, so far, that the increase in housefire risk is lower. But he is comparing the eight months of this year to the whole of 2008. Furthermore, I think, he has made the implicit assumption that shoddy workmanship will result in a housefire almost straight away and that the future stock of housefires will be unaffected by current shoddy workmanship. So we could be facing a housefire epidemic (not sure if that’s the correct term).
So where does this leave us? Possum could well be correct and people could well be over-reacting to a massive increase in the flow of insulation that would statistically give us an increase in housefires. Given their behaviour I don’t think the government believes this to be the case. They could have said this three weeks ago. I can’t believe that they couldn’t articulate this argument if they believed it to be true. Afterall they would have expected this to happen. Of course, that doesn’t change the fact that Possum could be right. On the other hand if, as I suspect, he has under-estimated the marginal effect of the new insulation on both the flow and the stock then we might have a massive increase in housefires due to this program over the next few years. Unless, of course, if an audit is undertaken of the work already undertaken and any problems fixed (at great expense).
Given the paucity of data I don’t know how Possum’s analysis could be improved upon to get a better feel for the numbers. But he is very handy at monte carlo analysis so it would be nice to see some scenario analysis using that technique.
