Rank-biserial and point biserial correlation coefficients in SPSS +

[BOAZ]

Hi there,

I have tow questions regarding these analyses that I am yet to find an answer for.

1) Rank-biserial
Is it possible to run this in SPSS? I have not found this option within the program and cannot find relevant advice on any forums. I am looking for relationships between sets of ordinal and dichotomous/nominal data.

2) Point Biserial
For a different analyses I am comparing scale vs. dichotomous data. My understanding is that a point biserial analyses may be achieved by running a standard pearson co-efficient test. However, my sample size is small (N = 17) and features some outliers. Although the data pass shapiro wilk normality tests, SW is weak for samples of this size, and the charts do not really appear to approximate normal data. Would you agree that the use of point biserial coefficients would be unsuitable for this reason?

Many thanks

 

[Dragan]

Hi there,

I have tow questions regarding these analyses that I am yet to find an answer for.

1) Rank-biserial
Is it possible to run this in SPSS? I have not found this option within the program and cannot find relevant advice on any forums. I am looking for relationships between sets of ordinal and dichotomous/nominal data.

2) Point Biserial
For a different analyses I am comparing scale vs. dichotomous data. My understanding is that a point biserial analyses may be achieved by running a standard pearson co-efficient test. However, my sample size is small (N = 17) and features some outliers. Although the data pass shapiro wilk normality tests, SW is weak for samples of this size, and the charts do not really appear to approximate normal data. Would you agree that the use of point biserial coefficients would be unsuitable for this reason?

Many thanks

You can use the Mann-Whitney test to address both of your concerns. For part 1, the Rank-biserial is just a linear function of the MW test. For part 2, the two-independent samples t-test will yield the same p-value as the point biserial correlation, thus, use the MW in lieu of the point-biserial correlation -- if non-normality is your concern.

 

[BOAZ]

Thanks for the advice Dragan. I hope you don't mind elaborating, but I haven't fully grasped your answer (my stats knowledge isn't that great!). For part 1, how do I derive a correlation co-efficient and p-value from the Mann-Whitney output? For part 2, I see that the p-value for the two-independent samples t-test matches the person p-value, but how do I obtain the co-efficient? A further problem is that the two sets of data are not independent, in that both were obtained from the same set of participants. Thanks again, B.

 

[Dragan]

A further problem is that the two sets of data are not independent, in that both were obtained from the same set of participants.

This new information blows your original post sky-high.

 

[jabbar.links]

Dear Dragan,
let say i have out put such as

Ranks
designation N Mean Rank Sum of Ranks
cgpa 1 4022 3928.95 15802257.00
2 3795 3887.85 14754396.00
Total 7817
then we use for rank biserial = 2*(3928.95-3885.85)/7817
= 0.010515543
for p-value we use this out put?

Test Statistics(a)
cgpa
Mann-Whitney U 7.551E6
Wilcoxon W 1.475E7
Z -.805
Asymp. Sig. (2-tailed) 0.421
a. Grouping Variable: desination

p-value = 0.421

so is it correct? please explain that this procedure is correct for the coefficient of biserial correlation and p-value??