Mann Whitney test

[Martin Imelda Borg]

Hi trying to do a Mann Whitney test - used the code as seen below:

ranksum Newtreadmill, by(Fitbit) porder exact

The following error came up:
more than 2 groups found, only 2 allowed
r(499);

I am only using 2 groups....


CODE]
* Example generated by -dataex-. For more info, type help dataex
clear
input float(Fitbit Newtreadmill)
645 602
658 636
717 692
691 709
730 654
end
[/CODE]

[Leonardo Guizzetti]

In your data example Fitbit takes on accettala distinct values. The by group Missy be coded as 0 and 1, two groups. Try an example from the manual to see this.

[Rich Goldstein]

try the following which first renames your variables so they can be used in -reshape-, then reshapes the data so it can be used for -ranksum-; note that I have not put labels on "count" but you can label them as Fitbit and as Newtreadmill:

. ren Fitbit ex1
r; t=0.13 9:52:23

. ren Newtreadmill ex2
r; t=0.00 9:52:34

. reshape long ex, i(id) j(count)
(j = 1 2)

Data                               Wide   ->   Long
-----------------------------------------------------------------------------
Number of observations                5   ->   10          
Number of variables                   3   ->   3           
j variable (2 values)                     ->   count
xij variables:
                                ex1 ex2   ->   ex
-----------------------------------------------------------------------------

. ranksum ex, by(count) porder

Two-sample Wilcoxon rank-sum (Mann–Whitney) test

       count |      Obs    Rank sum    Expected
-------------+---------------------------------
           1 |        5          33        27.5
           2 |        5          22        27.5
-------------+---------------------------------
    Combined |       10          55          55

Unadjusted variance       22.92
Adjustment for ties        0.00
                     ----------
Adjusted variance         22.92

H0: ex(count==1) = ex(count==2)
         z =  1.149
Prob > |z| = 0.2506
Exact prob = 0.3095

P{ex(count==1) > ex(count==2)} = 0.720

[Martin Imelda Borg]

Leonardo Guizzetti what do you mean by by group ‘Missy’ please ?

[Martin Imelda Borg]

Rich Goldstein apologies can you explain why you use ‘long ex, i (id) j(count) when you renamed the variables ax ex1 and ex2 - thanks
reshape long ex, i(id) j(count) (j = 1 2)

[Rich Goldstein]

it made it easy to use the command that I used to get the values in one variable (column) and the grouping variable in another - note from the ranksum help file that this is how the data need to be set up

[Leonardo Guizzetti]

Leonardo Guizzetti what do you mean by by group ‘Missy’ please ?

Oops, that was my autocorrect, and was meant to be Fitbit. My apologies. Rich has demonstrated this for you.