Some Frequently Asked (or just plain darn good) Questions

(New questions in RED at the top of the page)
 

•   Assignment #4, I am not sure that how I interpret my excel output. Do you mean to show it by calculation? If so, how?
  • For the week 4 assignment, you should comment on the sign of the estimated coefficients (are they what you expected?) and the magnitude of the coefficients. Does the regression result make sense? Plotting the data with the regression line can be helpful.
• I am very much confused whether or not my dependent variable is SA (seasonally adjusted). the content details say:
  
  NEW MOTOR VEHICLE SALES, MONTHLY, UNADJUSTED (RAW), IN UNITS (LEVEL 1) AND
  THOUSANDS OF DOLLARS (LEVEL 2). UNITS ONLY ARE ADJUSTED FOR SEASONALITY (SA).
  PROVINCIAL DATA IN UNITS AND THOUSANDS OF DOLLARS, UNADJUSTED (RAW)
  
  This series originates from a matrix which is has included both provincial and national data. However the description ion states clearly, again :PROVINCIAL DATA IN UNITS AND THOUSANDS OF DOLLARS, UNADJUSTED (RAW)
  • The content details are for the whole matrix - not for the individual series at hand, therefore the content details are virtually worthless. For unemployment rates, for instance, you have to look at:
    
    Contents:
    
    LFS CANADA CHARACTERISTICS MTHLY S.A. / UNEMPLOYMENT RATES 15 YRS & OVER (SA)
    
    This means it is seasonally adjusted, and
    
    Scaling Factor and Unit of Measurement:
    
    UNSCALED RATE
    
    means a simple percentage.
     
•   I'm at the end of my project now but I have one minor problem. I don't know how to look up my DW stat. My observations is 139 and K is 6 and the DW table at the back of the book only gives dL and dU for n up to 100 and K up to 5. what do I do, is there another way of looking it up?
  • Use the values on the table closest to your values, n=100, k=5.
           
•   In front of me is an Excel chart displaying the sample correlations between each of my variables. Since squaring the "r" is also equal to the coefficient of determination,
  • This is true only in the single regressor case.
     
  •   is it telling me that there may be multicollinearity between 2 variables if I find an "r" that's greater than my current "square root of my R squared"?
    
  • No. High correlations betw var.s is CONSISTENT w/ problems of multicollinearity, but does NOT mean there ARE problems of multicollinearity - problems are shown w/ t- and F-tests. The correlations are simply interesting to look at and report, to get a feel for your data.
          
•   I want to use the assign. #9 dummy variable (stockmarket crash). How can I explain it well?
  • I can't write the project for you. Try using demand theory to motivate it.
       
•   To calculate the Durbin-Watson, do we use just the "Residuals" or the "Standard Residuals" produced by the regression? I'm pretty sure I should just use the "Residuals" - however when I do that, I get an awfully high (like 5000 something) Durbin Watson statistic. When I use the "Standard Residuals" it's much lower (2.85 something). Both of these seem high compared to all examples found in the book.
  • The range for the DW is 0 to 4 - anything beyond means there is an error somewhere. If it is over 2, you have negative autocorrelation, which can happen.
     
•   Does it really matter if a data is either seasonally adjusted or not seasonally adjusted? And i don't quite know what is the difference between the 2 sets of data.
  • It doesn't matter for your project. The seasonally adjusted data has had seasonal effects removed - with dummy variables (and other more complicated methods).
     
•   When a data is adjusted, does it also mean the same thing as seasonally adjusted?
  • Not necessarily - - you are going to have to look over the data header info more carefully - if data has been seasonally adjusted, it will say "Seasonally Adjusted" or "SA". There are other adjustments possible (deflating data, scaling it into the thousands - so 14 means 14,000, etc.) you are going to have to read your header info and find out. If you can't understand it, drop by the computer lab and ask one of the lab TAs, or drop by my office hours.
         
•   In our project's intro, we're supposed to address our 'audience'. Who are they and how should we address them in the introduction?
  • You are supposed to decide who you are writing this report for - the auto manufacturers, or Consumer Reports, etc.
       
• If I include lagged variable, it will surely have influence on my dependent variable which gives a higher R-square. However, according to the text, the R-square resulted from such model is not many business and economic time series exhibit a rather smooth evolutionary pattern through time. Little attention should pay on the r-square in this kind of regression model. So, my R-square with lagged variable will not make much sense, right?
   
  • You still interpret the R-squared the usual way. It is true that the R-squared is not very useful for some economic data regressions, but it is very complicated to sort out when this is a problem. We will not discuss the complications you are referring to, so do not bother with them for your project.
• What's the difference between RAW data and Adjusted data?  Do I have to do something special to one of them when using it?
  • Well, one is adjusted - deflated for inflation, de-seasonalized, etc. - and the other (RAW) isn't. You should just mention if your data is adjusted in any way and how it is adjusted.
• Is it a problem for me to have a low R-square? I mean do I have to find independent variables in order to make my R-square bigger.... to be e.g 0.80 ?
  • No! I do not care if the R-square is high or low - I just care about your explanation of the results and your explanation of how you chose your model (the economic theory behind it).
• Will marks be deducted because of a low R-square?
  • No.
• When I compare to my original R-square to my new R-square with one extra variable, the new R-square is higher than that of the original. Does this mean I should add that new variable?
  • You should add variables only if there is economic theory to support that variable. The change it makes to the R-square does not matter.
• I want to use prices of the substitutes and complements for my dependent variable but all I could find is price indexes. What should I do about it? Is there a way to convert the price index to the actual price, if not, what should I include in my model then?
  • Price indices are fine.
• If some of my independent variables end up having the wrong coefficient sign then when I do my t-test, should I still do a one-sided alternative as what I'd expect to be right or do I perform a test that matches the way the data comes (even if it is contrary to economic theory of demand?) (i.e. my price of cars has a positive coefficient - when I'd expect a negative one - so should my alternative still be H1 < 0 which agrees with theory - or should it be H1 >0 since that's the way the regression turned out?)
  • Just comment of the obvious 1-sided alternative even if it is counter-intuitive and remark that this is against theory. An implication of this is either 1) the model is misspecified or 2) the theory is wrong. Comment on this in the project. writeup
• For our project, will you give a higher grade for those who get more useful explanatory variables? I mean, will  I get a better grade if I can find more suitable independent variables?
  • You will not be penalized for variables that don't work very well - are insignificant in the regression. Good interpretation and analysis is what counts.
• If I can think of independent variables but I can't find them, can I just tell you about this instead of adding them?
  • Yes.
• Is it necessary to add a lagged variables to our model?
  • It is not necessary to add a lagged dependent variable to your model - but it may be interesting to try it. Habit persistence of consumers can motivate this sort of variable.
• In the introduction part, you want us to outline the economic theory. I don't know how precise I should tell here 'coz we have to explain the theories one by one in part B.
  • State general theory in the introduction, then comment on your particular variables in part B
• How would i deflate i my data?
  • Divide all prices by the CPI for all goods, then multiply by 100. But this is NOT required for the projects (I don't lecture on price indices until the last 2 weeks of the course). If students do not deflate, they will NOT be penalized when the projects are graded.
• Why does the link to CANSIM not work sometimes?
  • If you are not on campus it will not work - only computers up on campus are permitted to link to that site. It will work from home, however, if you use SFU phone lines, and launch your browser after you connect to the campus.
• This question is one of those "true" or "false" questions: I say that the significance level of a test is the probability that the null hypothesis is true, is false.  I based my reasoning on the definition of the level of significance that the probability of rejecting null hypothesis that is true.
   
  • No - false. We start by assuming the null IS true. Then we ask,
    how often will our rule mislead us? When we use an alpha rule, our rule
    misleads us - has us rejecting the TRUE null alpha % of the time.