Hi! I hope somebody with greater expertise in statistics and spss than mine can give me a hand. I am conducting a pedestrian analysis from walking pace and number of people. I have 7 variables with 480 samples each and it happens only one of them is differently scaled from the other 6. So, my doubt is concerning whether it is necessary or not to apply any standardization method during a hierarchical cluster analysis on spss. The 6 same scaled variables describe walking pace (ranging aprox between 0.9 to 1.8 m/s) and the different one describes number of people (ranging on the hundreds).
Thanks in advance.
Hi Giancarlo,
Normalization (standardization, in SPSS) is recommended in general.
In your case, you could either do standardize the whole data, or get rid of the attribute the have different scale. Finally, one strategy would be to carry out the cluster analysis twice — once without standardizing and once with — to see how much difference, if any, this makes to the resulting clusters.
HTH.
Samer Sarsam, PhD.
Thank you for the soon replies Samer and Raid. @Samer, Thanks for the piece of advice. Just to add, which cluster method and interval do you recommend? would be ok with the default setting? ( Between-groups and squared euclidean distance) @Raid Amin. Thank you for the reply, as matter of fact, It is a multivariate cluster what i require
No worries. You'er welcome, Giancarlo.
The settings, in general, are based on your task. I'd recommend using the "Euclidean distance" in conjunction with "Nearest neighbour" method.
Cheers,
Samer Sarsam, PhD.
Thank you so much Samer for the clarification. My apologies, just one more thing I forgot to ask is, what Transformation Value do you recommend to use? Range -1 to 1?
No worries, Giancarlo.
For the transformation scale, it's usual to normalize all attribute values to lie between 0 and 1.
Cheers,
Samer Sarsam, PhD.
- Badges
- Science topic