# variability in measurement/reproducibility

#### statsnoob2022

##### New Member
I am looking to measure ultrasound cross sectional area (measured in cm^2) of a thigh muscle in people
I know that there might be some fluctuation in measurement/variability

Would this be a reasonable method to reduce variability: if i measure it once, call it "m1", then repeat the method immediately (m2). I then want to see if the 2 measurements are "close enough" - eg within 10% before accepting one of them as the "true value".
do I simply do m1-m2/m1 * 100 ?
but also doing m2-1/m2 * 100 would give slifghtly diff results

After I figure out this method, then I plan on testing it on a pilot study
- eg 10 participants, measure each participant twice. Then work out an intraclass coefficient, which I believe would be the Two-way mixed effects model?

Last edited:

#### katxt

##### Well-Known Member
Is the study to determine how accurate the ultrasound method is?
Why stop at two measurements? Is the ultrasound method expensive/painful/time consuming?

#### statsnoob2022

##### New Member
Is the study to determine how accurate the ultrasound method is?
Why stop at two measurements? Is the ultrasound method expensive/painful/time consuming?
Yes this pilot study is to see how "reliable/reproducible" the ultrasound method is.
The US method is quite time consuming (not particualrly painful).

#### katxt

##### Well-Known Member
do I simply do m1-m2/m1 * 100 ?
but also doing m2-1/m2 * 100 would give slifghtly diff results
A more common approach is the coefficient of variation = SD/mean which in the case of two measurements turns out to be
(m1-m2)/(m1+m2) doing the bigger mean - the smaller.
You are looking into "repeatability", same lab, same operator, same machine. You will probably also be interested in "reproducibility" different labs, different operators, different machines. This is a very common situation in comparing lab results and probably worth looking up.
Probably the analysis you mentioned is more complicated than you need.

#### Karabiner

##### TS Contributor
I then want to see if the 2 measurements are "close enough" - eg within 10%
Why a percentage? That would mean to accept a larger difference
when measuring larger areas. Wouldn't you rather be interested in
"a difference not larger than x cm^2"?

With kkind regards

Karabiner

#### statsnoob2022

##### New Member
Why a percentage? That would mean to accept a larger difference
when measuring larger areas. Wouldn't you rather be interested in
"a difference not larger than x cm^2"?

With kkind regards

Karabiner
because eventually the Ultrasound measurement will look at one participant in 2 time points, and we are interested in changes in size between the 2 time points - particularly participants who lose a % size (eg 10% size)
so the scannign method needs to be sensitive in picking up a certain % loss