I'm working with a longitudinal panel dataset that surveys the same few thousand individuals biennially and I have six years of this data. My dependent variable would be how much individuals spend on X, and there are a list of about 15 independent variables such as age, income, etc that could affect whether and how much individuals spend on X.

I could run regressions for each panel to see whether people who have higher incomes spend more on X, for example, but I want to be able to measure whether the spending patterns of individuals on X changes as their circumstances change (do individuals who get a raise between surveys spend more on X? do individuals spend more on X as they get older?) instead of just whether richer individuals spend more on X. I have no idea how to go about this, though.

What I have so far is that I could difference the data between years so I have (for example) IncomeChange and SpendingOnX for every year, and so on and combine everything as a set of entries, but that doesn't measure individual behavior changes as much as just the effect of changes. Such as:

..... | IncomeChange & all indep vars | dSpendingOnX

...1 | [2007 to 2005 for individual A].. | [how much more A spent in 2007]

...2 | [2009 to 2007 for individual A].. | [how much more A spent in 2009]

...3 | [2007 to 2005 for individual B].. | [how much more B spent in 2007]

...4 | [2009 to 2007 for individual B].. | [how much more B spent in 2009]

...5 | [2011 to 2009 for individual A].. | [how much more A spent in 2011]

And so on, down to entry 25,000 (for the same 5000 people surveyed each of the years, for 5 years since differencing wouldn't be available for the first year of the survey).

That would let me see how changes in, say, income, affect changes in spending in general, but I feel like there would be methods that would be a lot more insightful and I just haven't been able to figure them out. Would I have to run a separate regression for each year-combo (one for 2007 with differences in variables since 2005, one for 2009 with differences in variables since 2007, and so on?)

For example, with this level of information:

UserID | dIncome05to07 | dIncome07to09 | dIncome09to11 | dSpending05to07 | dSpending07to09 | dSpending09to11

A | +1000 | -3000 | +250 | +100 | -200 | +100

B | +2000 | -5000 | +700 | +400 | -100 | +100

...all the way down to the 5000th individual.

How would I be able to find results in the format of something like: "in general, as individuals see a change in income of +1000 the amount they spend on X is expected to change by +117.42"? Or is that not possible with this set of longitudinal panel data? With 5000 individuals, it's not like I could add, say, a dummy for each individual or something (to my knowledge) or subset them into subgroups.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.