Statistics 4

[25362]

Are statistics indeed invaluable as evidence in support of conclusions ?

 

[HVAC68]

A very generic question. The answer would be "it depends". Can you be a bit more specific.

HVAC68

 

[25362]

HVAC68.

For example, if a plant safety audit shows a yearly average of 10 dangerous occurrences (I refrain from using the word accidents to eliminate its connotation of chance), would a drop in that number to three be statistically significant ?

 

[PSE]

It might be an anomalous or random drop unless it stays at a new average of three for a while or if you could relate the drop to direct efforts to improve. For example if you did a pareto of incident types and worked to reduce the ones with greatest frequency, you could link those efforts to an effective drop in occurrences.

Regards,

 

[CajunCenturion]

Again, I think the answer is "it depends".

If for example, the plant was running three shifts and had an annual average of 10 occurences, then backed down to one shift and the annual occurences dropped to three, then I would say no, it is not statistically significant. If on the other hand, the plant was running one shift with an annual average of 10 occurences, then doubled production with a drop to three in occurences, then that would be significant.

There is too much of the big picture missing to know what the measurement means.

Good Luck
--------------
As a circle of light increases so does the circumference of darkness around it. - Albert Einstein

 

[BobPE]

stastics to the educated technical person are a guide to help understand things. Stastics to the lay person are scientific conclusions written in stone.

I was always tought that there are lies, damn lies, and stastics.

My answer would be that they are not valuable alone as evidence to support conclusions. They are only valuable in supporting the person that is authoring the conclusions.

Bob

 

[HVAC68]

I presume that your question is "all other things remaining same, does a reduction in the occurences from 10 to 3 mean anything ?" - Right ??

If that is so - then as somebody pointed earlier, one has to look for consistency in these reduction - did it happen overnight ? or did it reduce linearly over a period of time ? If it's a reduction happening in a consistently linear manner (maybe even exponential !!!) - i.e. as long as there is a pattern that proves that the reduction is happening consistently, then, it proves that action has been taken towards achieving it.

HVAC68

 

[HgTX]

They can be valuable when handled properly. In other words, a "this is more/less than that" is worthless, but showing degree of significance of the correlation, discussing possible factors that might lead to a noncausative correlation, etc., can be.

Numbers waved around are worthless. Used responsibly, they are a valuable tool.

Hg

Eng-Tips guidelines: faq731-376

 

[BobPE]

I would rearrange what you said HgTX, I hope you dont mind...

Numbers waved around are useless. The person using them responsibly is a valuable tool.

I am quite confident that I can make anything appear any way I want using statistics. It is how I use statistics that I will be judged by from my peers.

Bob

 

[SacreBleu]

I agree.
100% of people who eat bread die eventually.

 

[IRstuff]

Are statistics indeed invaluable as evidence in support of conclusions ?

Sure, when used correctly, with the correct assumptions and caveats.

For example, if a plant safety audit shows a yearly average of 10 dangerous occurrences (I refrain from using the word accidents to eliminate its connotation of chance), would a drop in that number to three be statistically significant ?

Not necessarily. It depends on the overall statistics. What is the variance of the historical values? A sigma of 3 would put the new datapoint well within the statistical distribution of the historical data. Until you can collect sufficient data to show that there is a "new" average around 3, you really can't make any conclusions.

TTFN

 

[dbuzz]

Statistics mean everything and nothing at the same time. One could almost always find a statistic of some sort to support one's point of view.

There is a great quote attributed to former British Prime Minister Benjamin Disraeli - "There are three kinds of lies: lies, damn lies, and statistics."

 

[IRstuff]

Spoken like the mathematical illiterate that he probably was.

Statistics, for all the uncertainty that comes with it, is a formal and rigorous discipline. No statistic is a lie; but there are plenty of liars about statistics. While some of you might argue that statistics is not engineering, engineering respect is based on the rigor of ones math. To cater to uneducated blather about statistics equated with damn lies from the mathematically challenged is to denigrate your own profession. Almost all aspects of engineering performance of materials, circuits, and systems are based on statistical behavior. I

f you think that statistics are lies and mean nothing, then you've basically laid claim that your own engineering analyses are so much horse manure; which they might very well be. For those who are PEs that believe that, they should immediately report themselves to their respective professional boards and declare themselves to be a danger to the public.

TTFN

 

[HgTX]

Wot he said.

Hg

Eng-Tips guidelines: faq731-376

 

[dbuzz]

Some commendable passion for statistics there IRStuff, however I think the Disraeli quote goes more to one's use of statistics than to statistics themselves.

One shouldn't blindly accept statistics without regard for the context. For example, take 25362's query: if a plant safety audit shows a yearly average of 10 dangerous occurrences, would a drop in that number to three be statistically significant?

What does the "drop" to three occurrences, from an average of 10, really mean? What is the standard deviation of the sample that produced the average of 10 occurrences? What if the average of 10 was produced from the following samples: 17, 5, 10, 1, 4, 30, 10, 4, 2 and 17?

Perhaps a better statistic is the frequency rate of occurrences (say per million man-hours), which is how safety statistics are usually reported.

 

[HVAC68]

That's what I said in my second post - assume all other things remain same - a drop in occurences would definitely mean something

HVAC68

 

[MRM]

We should take all this advice with a grain of salt; let's not forget that about 50% of all practicing engineers graduate in the bottom half of their class...
scary (shiver)

25362, your question is a good one. You're basically asking if your stats mean anything. Look into the t test, Student-Newman-Kuels(sp) test, and Duncan's range test. Some others might exist too.

These have been helpful for me, and they could be useful in helping you decide if your data set is a) large enough to give you statisically relevent results, and b) to help you decide if the change in number of incidents you're observing is statistically "different."

 

[CajunCenturion]

==> let's not forget that about 50% of all practicing engineers graduate in the bottom half of their class.

Not quite MRM. That assumes that all practicing engineers actually graduated, and further assumes that all who graduated are still practicing.

Good Luck
--------------
As a circle of light increases so does the circumference of darkness around it. - Albert Einstein

 

[rhodie]

Statistics never lie. Like computers, they only do what you tell them to do.

Inputs can be garbage.
Constraints can be applied with a bias.
Results can be interpreted improperly.

The usefulness of statistics rely entirely on the competence and comprehension of the user.

...and a general rule for interpreting trends or forecasting: the more data points, the better.

 

[BobPE]

could not have said it better rhodie!!!!!!!!

Bob