Stats Novice need help!

#1
Hello all,

I am in search of some guidance and direction. I work in the field of health and safety risk, forgive me if I sound completely uneducated in the subject of stats, because I am.

I am looking for some statistical methods to provide my work some predictive analysis on health and safety risks... So if I put it in lay terms, I want to look at things such as historical incident and injury statistics and other forms of subjective data like risk assessments on a heat map to provide some kind of prediction data to inform the leaders of what "could" be coming.

So I am not after anyone to teach me the "how" just some suggestions that I can go away and learn on my own.

e.g. "Use Murphy's statistical law of screw ups - it provides you a way to predict non numerical and subjective data outcomes"

Any ideas is much appreciated!
M
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
The first thing to always do is visualize your data and look at descriptive stats. Then start reading entry level epidemiology then biostatistics books. Also read relevant journal article on the context/content you are interested in to learn how others conduct similar initiatives.
 
#4
Thanks for that!
I have just started watching about 3 hours of biostatistics. Very helpful!

Is there any methods or suggestions for handling data that isn't quantifiable?

For example in my situation if I'm looking at data of incidents I can see how many "forklift" incidents have happened over a period. But I would like to analyse the severity and that is measured by non numerical values... first aid treatment, medical treatment, permanent disability, fatality... terms like that...

So I could have 12 first aid forklift incidents.. 25 medical treatment... 3 perm disability and 1 fatality... if I wanted to look at the probability would I assign each a numerical value? And how would I scale it? Is fatality 2x permanent disability?

Thoughts?
 

Karabiner

TS Contributor
#5
You could consider it as an ordinal scale, i.e. the consequences ordered
none - first aid - med treatment - permanent disability - fatality. There
are techniques for analysing such ordinal scaled variables But
admittedly I did not quite exactelly understand what you are
trying to achieve.

With kind regards

Karabiner
 
#6
You could consider it as an ordinal scale, i.e. the consequences ordered
none - first aid - med treatment - permanent disability - fatality. There
are techniques for analysing such ordinal scaled variables But
admittedly I did not quite exactelly understand what you are
trying to achieve.

With kind regards

Karabiner
Hey @Karabiner

I will need to look into that ordinal scale.

I will give you a scenario of how I am trying to apply this... first I will provide some parameters that we use in risk to rate a risk (typically 5x5 matrix)

Safety Risk Outcome Descriptors (low-high)
No injury/First aid - Medical Treatment - Temp Disability - Perm Disability - Single/multi Fatality

Likelihood of Occurrence Descriptors (low-high)
Monthly - Yearly - 5 Yearly - 30 yearly - 100 yearly

If I took a hazard like a chainsaw. there are a few things I need to consider. First is what is the potential range of outcomes that are possible from an incident with a chainsaw. In this case a chainsaw has the potential to result in an injury from No treatment->Fatality so it covers all options, but for example a staplers range would probably only be No treatment -> Temp Disability. VS what incident outcomes have actually happened.

Then I would like to show...

What is the probability of the number of chainsaw incidents for next year: this one is easy because I can find the mean of previous years data assuming there ha been no operational changes (eg. we haven't decided to buy or become arborists)

What is the probability of injury outcome should an incident occur: This one is much harder because data is limited, I would say for certain we have no incidents of fatality in known record. But that doesn't mean that the probability doesn't exist.

So I guess what im trying to wrap my head around is... Lets say the data says I had 10 incidents last year and all 10 were temp disability... Would my standard deviation be.... First aid<---2 Med Treatment<---1 Temp Disability(0) -->+1Perm Disability -->+2 Fatality ? I don't think I am sold on it being this because it hard to assign a "value" to the difference between Perm Disability and fatality. Is it twice as unlikely? 10x as unlikely? 1000x?

Thoughts?
M
 
Last edited: