2008 Bk2 #2: it says that if you spot trend in the plotted data (upward or downward slope), you compute the first differences of the data. But on page 223 it also says that the log-linear model is best for a data series that exhibits a trend. My question: since you would use the log-linear model, why would you need to compute the first differences? The next bullet point says that if the data plot in a curve, you must use a log-linear using first differences. My question: Figure 1 on page 223 simply use log-linear trend model to transform the curve with out using first differences. Can somebody please get this straight for me? what’s are the rules? Thanks.

Turn to the back of the book, it gives you a break down on the steps you must take to determine if your model is correct. Your first question is: Time series or regression If you choose time series, plot the data and determine if it is a linear trend or a exponetial trend. linear trend, simple linear model exponential trend, log-linear model your first differencing comes in when your data isn’t covarience stationary.

kerry1 Did some further reading. I think it means: 1.if you spot upward or downward trend on the raw data à not covariance stationary à first differencing à use the linear trend model 2.if you spot a curve (exponential growth) à not covariance stationary à first differencing à use the log-linear trend model I’m not sure about #2 if you should do log-linear first and then first differencing.

its log linear first.

log linear, then first differencing, then linear or log linear model?