0
$\begingroup$

I have some (2749) sensor output values. They range from 0 to 1769. With google spreadsheet I put it thru some calculations to get the (mean) average and standard deviation. With this I created two columns BINS and NORMDIST to create a chart. Please look at the attached image.

Mean is 53.35. Standard deviation is 61.23

What can I extrapolate from this?

A) Is one deviation 61.23, two deviation 122,46? Two deviations gives me ~95%.

B) So since my mean is 53, 95% is a effective range from 0 to 122 (as I don't have negative data)?

My chart shows while the range is from 0 to 1769, its mean is 53.

C) Is my data bad? The form of the curve the data makes is very "typical" distribution, but a little skewed to the left?

enter image description here

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ It's (very) right skewed. Left or right: which term to use depends on which tail is longer on a conventional histogram (or similar graph with magnitude axis horizontal). The right tail is longer. (Even a few texts get this wrong, and it's a tell-tale sign of incompetent authors.) $\endgroup$
    – Nick Cox
    Commented Nov 26, 2018 at 11:59

1 Answer 1

Reset to default
3
$\begingroup$

Welcome.

For A) No, the rule about 95% within 2 standard deviations is for normal distributions. You haven't got a normally distributed variable. For one thing, your variable has a lower bound. It also seems to have a lot of skew.

B) No, for the same reason as A.

C) Is it "bad"? Well, maybe it is and maybe it isn't, but there's nothing in what you've posted that says "this data is bad": it simply isn't normally distributed. Many variables are not.

Improve this answer
$\endgroup$
12
  • $\begingroup$ Thanks Peter. So since my data (from the chart) isnt normally distributed then I cant use standard deviation to say if I got an result that it is confined within the x% normal standard deviation? I just started working with my data and want to learn about the takeaway $\endgroup$
    – fUrious
    Commented Nov 26, 2018 at 11:58
  • 1
    $\begingroup$ You can always count how many of your values fall within mean $\pm$ so many SD. Just don't expect results for the normal distribution to hold for your data. $\endgroup$
    – Nick Cox
    Commented Nov 26, 2018 at 12:00
  • $\begingroup$ That's correct. It's possible to construct distributions where very strange things happen. $\endgroup$
    – Peter Flom
    Commented Nov 26, 2018 at 12:38
  • $\begingroup$ @NickCox 1710 of 1769 (~97%) has a value between 0 and mean (53). So since my data isnt normally distributed, this is the closest I get by counting check when I hit a level of certainty that I'm comfortable with $\endgroup$
    – fUrious
    Commented Nov 26, 2018 at 12:44
  • 1
    $\begingroup$ I can't easily comment on what you're comfortable with. There many possibilities for analysing these data. Why do you think it's important whether they are normally distributed, as they clearly aren't? $\endgroup$
    – Nick Cox
    Commented Nov 26, 2018 at 12:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.