I often look at stats particularly benchmarking stats with a cautious eye. You will often see incident rates compared and one agency will look much lower that the others and it is not obvious they are doing anything different. You just suspect that they are under reporting.

Here’s the thing, setting targets and looking for statistical measures of success is no bad thing – provided we are measuring the right thing in the first place. Too often safety stats (and I suspect crime figures) simply look at what has happened, i.e. number of near misses, number of LTA’s, in some cases days off sick etc. Then a target is set by someone to “drive improvement” and so the challenge is set. This gets even more complex when we incentivise someone to achieve that new target with bonuses, pay rise, Christmas hamper. Humans will behave in a human way.

So, if we see that 10 people slipped on loose gravel in the car park last year we might set a target of say 5 slips this year(work with me here – lets NOT go down the zero accidents route for a moment). As we approach 5 this year we become more worried. At 5 or 6 we start to question “Was that one really a slip”, “that one wasn’t due to gravel, it was high heels” etc.

If we were to set another target, alongside the reduced trips we might get a “better” result. So in my hypothetical scenario above how’s about setting the 5 trips target but also set a target to sweep the car park weekly. There can then be a direct correlation between “did someone slip and had the car park been swept that week?).

In other words, safety stats tend to be reactive - bad things that have already happened. By also measuring a control and reporting it as a safety stat we now have a comparative leading safety stat. And from that we can again incentives people to reduce the stats by measuring the input as well as the output.

There is another output – if the car park is swept weekly and the slips do not reduce there’s another problem, conversely it may turn out that there are no slips this year so you have proof your control worked for the next time the bean counters try to stop you sweeping the car park every week.

Jim

It is quite reasonable to be circumspect about policy targets at national level, unless they are validated by professional statisticians such as those employed by the Office for National Statistics, the HSE and (some) professional societies.

Within an organisation, the value of such figures depends on the skill level of those who gather and interpret statistical data. Far more useful analytical guidance can be provided to managers (and employee representatives) if statistical trends about safety are subjected to analysis using techniques such as multiple regression, correlation and cluster analysis, which can provide robust arguments for investing in training, PPE, new or refurbished furniture and tools, and similar kinds of interventions.

Competent use of inferential statistics are normally only referred to as 'lies, damned lies an statistics' by those without the understanding and knowhow necessary to recognise the value of interval estimates and levels of confidence in professionally-informed arguments.

Too often, they bedevil any debate about unavoidably complex issues such as behavioural safety, longitudinal accident/incident analysis, ergonomic interventions and safety culture by demonstrating - even loudly proclaiming - their profound lack of relevant statistical competence, even on the hallowed IOSH forums on occasions.

I have visited many businesses and although many try to keep their accident/incident investigations honest and try to gain a good undestanding of what is happening on site so they can produce good preventative actions I have to say I have also come across a few who literally throw accident and incident reports in the bin. I can think of two companies immediately that claimed zero accidents or incidents but had a knack of throwing away anything that came in. It is a poor management tactic of course.

Chris, very profound and true - nice post.