Help interpreting second differences

[NicV]

Dear all,

Panel unit root testing shows me that most of my variables suffer from non-stationarity, and are first-difference stationary. A couple of my variables however are second-difference stationary.
In have included a screenshot below for reference. Here you can see that I have twice differenced log(poptot) and log(popurb) while the rest of my variables are first differenced.
I interpret the coefficients of the first differenced variables as follows: e.g. if log(infmort) moves in an upwards direction in the current period, so does log(u5mort). A positive change in previous period's log(10+netoda(-1)) will cause a negative change in this period's log(u5mort) (up to a certain height after which it will positively change this period's log(u5mort) as demonstrated by log(10+netoda(-1))^2).
I hope I am correct in this.

Then how should I interpret the second differences? If the rate of growth of log(poptot) increases this period, it has a negative effect on the change of log(u5mort) in this period? So only if log(poptot) increases faster this period than it did in the last will it bring about a change in u5mort?

Help is greatly appreciated, and I hope my question isn't too unclear.

Thanks in advance!

Nic

[wariararachel]

i wouldn't go as far as working with second difference as this is not logically interpretable. we get percentage change with the first difference of the log. i recomend to use it then carry out a cointegration test. it says that if your variables have a long run relationship- this will be confirmed by the cointegration test, then you can work with non stationary series.