Finding the skewness and Kurtosis of grouped data

  • Thread starter Thread starter chwala
  • Start date Start date
  • Tags Tags
    Data
Click For Summary

Homework Help Overview

The discussion revolves around calculating skewness and kurtosis for grouped data. Participants are exploring the correct methodologies and formulas for these statistical measures, particularly in the context of using software like SPSS and PSPP.

Discussion Character

  • Exploratory, Conceptual clarification, Mathematical reasoning, Assumption checking

Approaches and Questions Raised

  • Participants are sharing their calculations and questioning the accuracy of their variance and skewness results. There is discussion about different formulas for skewness and kurtosis, including those used in SPSS and PSPP. Some participants are comparing their results with software outputs and discussing potential discrepancies.

Discussion Status

There is an active exchange of ideas, with participants providing insights into different methods for calculating skewness and kurtosis. Some have offered clarifications on the definitions and formulas, while others are reassessing their calculations based on feedback. Multiple interpretations of the formulas are being explored, and participants are encouraged to verify their methods.

Contextual Notes

Participants are working with grouped data and are mindful of the differences between population and sample statistics. There are mentions of specific software outputs and the need to understand the underlying assumptions in their calculations. Some participants express uncertainty about the definitions and methods, indicating a need for further clarification.

chwala
Gold Member
Messages
2,828
Reaction score
425
Homework Statement
See attached below
Note that this is an original problem created by myself.
Relevant Equations
Skewness and kurtosis- Statistics
See the grouped data below; I just want to be certain that i have followed the correct step in trying to find skewness of the grouped data.

1638258802517.png


1638258853711.png
 
Physics news on Phys.org
Your calculation of the variance is slightly off; it is correct for the variance of a population but not for a sample - you need to multiply it by n/(n-1). (Compare your definitions of Var and S.)
 
Ok i will check on that...on a side note am trying to also self learn spss; i tried to feed the information as per the table below;

1638280490837.png


and the results were as follows;

1638280560348.png


looks like my calculation of skewness was on point.
 
Last edited:
Well, the same to 2 sig figs (which may be perfectly adequate for your purposes). There are several slightly different formulas for sample skewness (see the Wikipedia article on skewness) and yours is not identical to any of them. It says that the one it calls G1 is the one used in SPSS, and I calculate its value as -0.1824 - very similar to yours but not identical.
 
I have just checked; The sample variance is now ok fully understood;

1638282134679.png
 
mjc123 said:
Well, the same to 2 sig figs (which may be perfectly adequate for your purposes). There are several slightly different formulas for sample skewness (see the Wikipedia article on skewness) and yours is not identical to any of them. It says that the one it calls G1 is the one used in SPSS, and I calculate its value as -0.1824 - very similar to yours but not identical.
Yes, i saw something like that, Peason's measures on skewness...using median, mode...which is the general accepted method world wide? And for kurtosis, which method is preffered? I will try and calculate kurtosis tomorrow...but hey, i do like the two methods that i have indicated...i think i will stick with that...:cool:
 
I am now trying to calculate Kurtosis, allow me to share an example that draws my point of reference;
This is an example from my research;

1638361764895.png

1638361790176.png


Now having shown this, and going back to our problem, it follows that, Kurtosis, ##K## for a sample population is given by,

##K##=##\frac {\sum(x- μ)^4}{ (n-1)s^4}##=##\frac{1,024,059.383+ 472,291.898+16,607.53125+1,561.929+215,755.34+876,623.45}{39×12.89^4}##

##K=\frac{2,606,899.531}{1,076,654.293}##=##2.42##

But checking this with spss, i am getting ##K= -##0.42##,

1638363008576.png


where could the mistake be?
 
Last edited:
Marks
x​
ffx(x-X)f(x-X)^2 =mf(x-X)^3
1-105.5211-26.751,431.13-38,282.59
11-2015.5693-16.751,683.38-28,196.53
21-3025.58204-6.75364.5-2,460.38
31-4035.5144973.25147.88480.59
41-5045.57318.513.251,228.9316,283.42
51-6055.53166.523.251,621.69
37,704.23​
4012906477.51-14,471.26

Also looking at this and calculating skewness, i am getting,

Skewness= ##\frac{-14,471.26}{39 ×12.89^3}##= ##-0.173##

and not ##-0.18## as we had expected...
Method 2; Using joanes and gills approach

Let n=##\sum (x-μ)^4= 2,606,899.531## then;
kurtosis = ##\frac {n}{m^2}## & excess kurtosis = ##n-3##

kutosis = ##\frac {2,606,899.531}{6,477.51^2}##= ##\frac {2,606,899.531}{41,958,135.8}##=##0.062##
excess kurtosis = ##0.062-3##=##-2.938##
then the sample excess kurtosis= ##\frac{40-1}{38×37}##×##[41×-2.938+6]##=0.02773×-114.458=-##3.17##
 
Last edited:
Skewness: I think this comes from using your formula with the correct value of S. As I said, your formula is not the same as that used by SPSS; the fact that you previously got (using your formula with the wrong value of S) the same value as SPSS (to 2 sig figs at least) was a coincidence.

Kurtosis: again there are several definitions. See wikipedia page; the quantity used by SPSS is the one they call G2. (You could also check the help for the KURT function in Excel.) Note that this is a formula for excess kurtosis = kurtosis - 3. In your second method, note that m = (n-1)s2, so your formula is a factor of (n-1) too small. (By the way, too many n's meaning different things!)
 
  • Like
Likes   Reactions: chwala
  • #10
Actually i am using PSPP which assumably, ought not be different from SPSS. I will dedicate this day to studying Kurtosis:cool::cool:...weighted tails...cheers mate
 
  • #11
chwala said:
Marks
x​
ffx(x-X)f(x-X)^2 =mf(x-X)^3
1-105.5211-26.751,431.13-38,282.59
11-2015.5693-16.751,683.38-28,196.53
21-3025.58204-6.75364.5-2,460.38
31-4035.5144973.25147.88480.59
41-5045.57318.513.251,228.9316,283.42
51-6055.53166.523.251,621.69
37,704.23​
4012906477.51-14,471.26

Also looking at this and calculating skewness, i am getting,

Skewness= ##\frac{-14,471.26}{39 ×12.89^3}##= ##-0.173##

and not ##-0.18## as we had expected...
Method 2; Using joanes and gills approach

Let n=##\sum (x-μ)^4= 2,606,899.531## then;
kurtosis = ##\frac {n}{m^2}## & excess kurtosis = ##n-3##

kutosis = ##\frac {2,606,899.531}{6,477.51^2}##= ##\frac {2,606,899.531}{41,958,135.8}##=##0.062##
excess kurtosis = ##0.062-3##=##-2.938##
then the sample excess kurtosis= ##\frac{40-1}{38×37}##×##[41×-2.938+6]##=0.02773×-114.458=-##3.17##
I have read on a better way of finding or rather measuring kurtosis and skewness. This is by using moments...quite straightforward. I need clarity on the highlighted part...

In general,
1639985641932.png

1639985677509.png


1639985747343.png


##d##=##\frac {(ΔMid-point)}{20}##
Now, we need to establish ##d## values from the class intervals. My point is that the textbook has a mistake on the highlighted part of the class interval...it ought to be, ##100- 120, 120-140, 140-160##... this is the part that i need clarity on, otherwise the other steps of the working are clear. i.e...
1639986246549.png


1639986275660.png


Therefore,

##β_1## and ##β_2## =

1639986387276.png
implying a 'symmetrical distribution' as the skewness is almost zero and a leptokurtic distribution.

I think the errors are typo...
in the last part, we should have;
##β_1##=##\frac{(μ_3)^2}{(μ_2)^2}##...
 
Last edited:
  • #12
chwala said:
Homework Statement:: See attached below
Note that this is an original problem created by myself.
Relevant Equations:: Skewness and kurtosis- Statistics

See the grouped data below; I just want to be certain that i have followed the correct step in trying to find skewness of the grouped data.

View attachment 293330

View attachment 293331
Now i will go ahead and calculate the skewness and kurtosis of this data...Give me a moment. Our table will look like this;
MarksFrequency##x_i####d####fd####fd^2####fd^3####fd^4##
1-1025.5-3.33-6.6622.178-73.852245.927
11-20615.5-2.22-13.3229.57-65.646145.735
21-30825.5-1.11-8.889.857-10.94112.145
31-401435.500000
41-50745.51.117.778.6259.57310.626
51-60355.52.226.6614.78532.82372.867
-14.4385.015-108.043487.103

I hope i am doing it right...i have finished my calculations and am getting different values i.e
##β_1=4.98## and ##β_2=2.48##...i may need to re check my calculations and at same time i would like to know on whether my table of values is correctly done...

A rough check on the Mean = Assumed mean + [##\frac{fd}{f}]×i##
= ##35.5##+##\frac{-14.43}{40}×9## =##35.5+-3.24675##=##32.25## which agrees with what we had found earlier in post ##1##
I may need to check moments ##2,3## and ##4## later...
 
Last edited:
  • #13
Hello again,

I don't agree with your calculation for the ##d##. You want to use the distance between the bin centers as denominator, i.e. 10, not 9. Think of the marks as a continuous distribution, or else you get gaps.

##\ ##
 
  • Like
Likes   Reactions: chwala
  • #14
BvU said:
Hello again,

I don't agree with your calculation for the ##d##. You want to use the distance between the bin centers as denominator, i.e. 10, not 9. Think of the marks as a continuous distribution, or else you get gaps.
Bvu nice talking again, I'll amend my excel and recalculate everything again...means for continously data we are thinking along the lines of [##0.5##( lower class limit) is less than or equal to ##x## <10.5]... to have our class interval as ##10##...cheers
 
  • #15
BvU said:
Hello again,

I don't agree with your calculation for the ##d##. You want to use the distance between the bin centers as denominator, i.e. 10, not 9. Think of the marks as a continuous distribution, or else you get gaps.

##\ ##
Just a minute what about ##x_i##? Will i now use ##5## instead of ##5.5## using the first class interval ##[0.5≤x<10.5]##as my point of reference?
 
  • #16
I do not think it would be possible to calculate skewness and kurtosis because of the discontinuity at the class intervals...I may need to create another suitable example...
 

Similar threads

Chi-squared test for normality
  • · Replies 14 ·
Replies
14
Views
2K
What is the derivative of a skew symmetric matrix?
  • · Replies 4 ·
Replies
4
Views
3K
Determine the type of correlation for the two variables given in the data
  • · Replies 4 ·
Replies
4
Views
4K
Plotting a Scatter Diagram from a Large Data Set
  • · Replies 4 ·
Replies
4
Views
2K
Can Nonlinear Equation Fitting Yield Reliable Parameters for Varying Data?
  • · Replies 6 ·
Replies
6
Views
1K
High School How is the harmonic mean affected by additional data points?
  • · Replies 4 ·
Replies
4
Views
2K
Undergrad Comparing Skewness and Kurtosis Levels: A Question for Data Analysis
  • · Replies 1 ·
Replies
1
Views
3K
Impact of several variables on resulting projectile motion trajectory
  • · Replies 3 ·
Replies
3
Views
2K
Undergrad Help me understand skewness in QQ-plots please
  • · Replies 1 ·
Replies
1
Views
2K
Probability distribution for discrete data
  • · Replies 8 ·
Replies
8
Views
2K