After doing research on different goods, I decided to examine the relationship between cigarette consumption and disposable personal income.  For my model, I gathered data on real disposable personal income per capita from FRED and cigarette consumption per capita from the US Department of Agriculture.  The data on real disposable personal income per capita was measured in terms of dollars, whereas cigarette consumption per capita was measured in terms of the number of cigarettes consumed.  I purposefully searched for cigarette consumption data that was measured this particular way because I felt it would be more digestible and applicable to the average consumer.  On top of making the data applicable, I also decided to run a time series model, instead of a cross section model.  I believed that a time series model would give a more accurate depiction of how cigarette consumption would response to changes real disposable personal income.

When creating my model, I selected cigarette consumption as my dependent variable and real disposable personal income as my independent variable.  Before beginning this assignment, I believed cigarettes to be a normal good, but I was wrong.  After running my model, I found that cigarettes were an inferior good, meaning that the amount of cigarettes consumed decreased as real disposable personal income increased.  As you can see from the parameter estimates below, if real disposable personal income (per capita) increased by one dollar, cigarette consumption (per capita) would decrease by approximately 0.162 cigarettes.  Although this decrease is small and to some might seem like an insignificant amount, when examining the changes in cigarette consumption overtime on an Engel Curve, like the one created below, it is easy to see how much a decrease of 0.162 cigarettes per dollar can have on overall cigarette consumption in the long run.





After running the model and examining my data, I came to the conclusion that my time series model had positive autocorrelation.  Positive autocorrelation did affect my data and the parameter estimators, but not by much.  If you examine the results from when I ran the model with no correction and then when correcting for autocorrelation, you will see slight differences in RDPI’s coefficient and R².  In model with no correction, RDPI’s coefficient was -0.16282 and in the model correcting for autocorrelation, RDPI’s coefficient is -0.1431.  Along with changes in RDPI’s coefficient, I also saw changes in R².  In the model with no corrections, R² was 0.8399, but when the model corrected for autocorrelation, R² was 0.951.

Although the model was very insightful and helped me determine that cigarettes are in fact an inferior good in today’s market, it would have been interesting to see how the model would change if I included more variables.  Another possible area of issue within the model comes from the years the data was collected—education on the deadly effects of cigarettes are well known and wide spread today, however back in the early to mid-1900s, this information was not available like it is today.  If it was possible, it would be very interesting to compare today’s cigarette consumption to the early/mid 1900s cigarette consumption in relation to real disposable personal income.


Does being rich mean you smoke more?