1

My data is reported at the household level (i.e. electricity consumption for the whole house) but it also has some descriptive data of the individuals within the household (i.e. there are 2 individuals in household A, aged 15 and 45).

For example:

enter image description here

What is the best method do determine what impact 15 year olds have on electricity consumption? Or similar question?

My initial approach was create some categorical variables with age bins and then estimate their impact using a linear regression. But I'm eager to hear of plausible alternative methods, perhaps a states based alternative?

Thanks for the help!

lexkel
  • 13
  • 2
  • You want to estimate the impact of 15 year olds or of age generally? – Heteroskedastic Jim Jul 06 '17 at 02:42
  • Probably either ultimately, i just need to estimate the relationship to incorporate into a demographic model. – lexkel Jul 06 '17 at 02:49
  • There is no variability within household on the outcome, as your outcome is at the household level. You would not run this as a multilevel model. I've given an answer on how you could test the effect of 15 year olds. – Heteroskedastic Jim Jul 06 '17 at 02:53

1 Answers1

1

Actually, you would not run this as a multilevel model, as your outcome is at the household level. It is a regular regression. There are a number of ways you could code each household:

  • Whether or not it has a 15 year old in it - binary variable
  • Number of 15 year old members of household - count variable
  • Proportion of household that is 15 years old

The first option reduces to a t-test. The second and third options will probably give you zero-inflated predictors. See the comments on this post for options to deal with zero-inflated predictors - "Zero-inflated" predictors in regression?

If you intend to represent other age categories as predictors, they can be added into the model using any of the options above; of course, be consistent.

Heteroskedastic Jim
  • 4,567
  • 1
  • 10
  • 32