1
$\begingroup$

I'm trying to compare if the order in col 1 is comparable to the order in col 2. The simplest thing I can do is just to compare both ranks using something like Friedman's test.

However, Col1 and Col2 is on a continuous basis. And for instance, the .94 and .93 in column 2 are closer together than 13.06 is to .94. Thus it's not quite ordinal data.

What would be a good way to compare the order but while taking into account the magnitude?

Thanks!

           Col1         Col2
7          1610635       1.61
8          3214571       2.04
9          2188633       1.30
4          1585069      13.06
5          2009116      14.74
1          2091101      12.50
2          2422071      14.64
3          1633422      13.05
10         3428149       0.93
11         2001198       1.20
12         2191020       0.94
13         2294395       8.83
14         2921134       9.18
15         2659132       6.60
  • Background information:

As requested, here's the background. Column 1 represents the overall counts of a particular type of T-cell using one type of instrument which reads raw counts. Column 2 represents the number of t-cells found using a different type of instrument which provides things on a percentage basis.

We are looking to compare that the two instruments roughly act the same -- they should pick up roughly the same number (rather on a percentage or on a raw count).

I don't have any other data -- just these two columns.

Thanks!

$\endgroup$
  • $\begingroup$ Can you amplify your question a little bit to motivate the research question you would like answered? Given that these seem to be very different measures, it is hard to know what kind of conclusion you are trying to draw (e.g. apples to oranges). $\endgroup$ – Alexis Aug 29 '14 at 21:14
  • $\begingroup$ @Alexis: No problem, added. $\endgroup$ – user1357015 Aug 29 '14 at 21:27
  • $\begingroup$ Why don't you make a scatterplot of these columns? They appear to be unrelated--in fact, they are negatively correlated! Unless one of these instruments is producing garbage, I am inclined to suspect the data were corrupted in the process of pasting them into this question. Some useful information may be found in the closely related thread at stats.stackexchange.com/questions/52572. More is available by searching "chemometrics". $\endgroup$ – whuber Aug 29 '14 at 21:36
  • $\begingroup$ @whuber: the plot is a good suggestion and I have. I'd like to do a correlation, but I'm somewhat hesitant of doing a pearson's correlation with such a wide scale. $\endgroup$ – user1357015 Aug 29 '14 at 21:43
  • 1
    $\begingroup$ If the second is a percentage, why not similarly normalize the first? $\endgroup$ – Glen_b -Reinstate Monica Aug 30 '14 at 0:57
1
$\begingroup$

I agree without more information it is difficult to assess what method might help you out. What is your reasoning for wanting to use Friedman's test?

What you can try to do is center/normalize your data. To center your data: Calculate the mean of each column, subtract the corresponding mean from each data point according to column. To normalize: divide by the number of data points. Then proceed to compare.

Again, without knowing what you are trying to achieve is difficult to provide an adequate answer.

Edit because of extra information:

You should either compare raw data to raw data or percentage to percentage. That is transform either one column either to raw or percentage. Then it will be easier to compare. Perhaps, you could do a difference test. For a graph just center and normalize the data.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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