stata 17 advice on how to create PCA asset measurement

[Chloe Williamson]

Hi! I am trying to run principal component analysis on my data set which includes 35 different asset types and how many of each is owned by a household. I would like to create a pca wealth indicator for each household based on the type of assets owned, the value of each asset (ie. how common is it to own) and the number of each asset owned. I have done some research as to how to create this indicator but I'm still a little confused and was wondering if anyone could offer any help as to how I would code this (and also how many components I would need to retain in order to get the most accurate measure). for an indication of the basic set up of my data set I include the following (the real data set however has 35 different asset types) :
household id stove fridge wardrobe water pump
1 2 0 4 0
2 0 1 3 2
3 4 0 1 1

I am using Stata 17.

Thank you for any help

[Nick Cox]

If you trust PCA here, the only answer I've ever understood is that the first PC is the best single summary you can find.

There is a meme or myth going round that somehow you can improve on PC1 by mushing it together with other leading PCs. I've challenged people pushing this to explain why or to give a reputable source and never that I can recall got an answer other than "I've seen papers doing this". This is to repeat the challenge to any and all readers.

Books on PCA are written by those who think it's a wonderful technique, which is not surprising, but I teach it for about 15 minutes once a year and I'm always on the lookout for convincing examples and am repeatedly frustrated. The best are when all the variables are really close measures of the same construct. Trusting it to somehow distill any complicated pattern of correlations to one magic summary doesn't often work well.

[Chloe Williamson]

Okay thank you so much for your advice!