Estimating the effect on office rents in a district when a certain office opens (running into issues)

[Waterloo RE]

Hi,

I have a few problems (probably very basic) that I could use some help with. I am trying to estimate the effect of a dummy variable (the opening of a particular type of office provider) having an effect on office rents in different districts. This is a uni project, and it is designed to be set up as a feasibility study (but I would like to see some results). I am using fixed effects panel data to estimate the effect on the growth in rents.

I have found a similar paper that used a methodology I would like to follow. They used the log difference of office rents (I've attached an image of their estimating equation below). My first issue is that some of my rent growth is negative, and hence I can't use this as a log variable. Do you know of anyway around this so that my analysis follows a similar technique? I am totally clueless about how they did the log difference of their employment variable (b3 below)since I have the similar data, and the difference between the previous period is negative in many cases.

This is my first attempt at a research project using stata (or any econometric method) so I am struggling at almost every stage.

I was thinking I could use a dummy variable for two to four periods before the office opened, and up to two periods after, but my friend said that my explanatory variables should only be for the periods before. He wasn't very clear, does this make sense to anyone?

Just read the FAQ on asking questions, example code below.

* Example generated by -dataex-. To install: ssc install dataex
clear
input double RentGrowth float WeWorkDummy
  .00245914631523192 0
  .00668658642098308 0
  -.0512353368103504 0
  -.0711766481399536 0
   -.050842959433794 0
  -.0234194714576006 0
  -.0131115764379501 0
   -.015505138784647 0
-.000390754081308842 0
  .00551744690164924 0
  .00869508367031813 0
   .0234992895275354 1
   .0308510307222605 0
   .0361947305500507 0
   .0242324471473694 0
   .0334946326911449 0
   .0377343520522118 0
    .034952737390995 0
   .0401232093572617 0
   .0336563810706139 0
   .0240535959601402 0
   .0297666415572166 0
   .0239342022687197 0
  .00722300400957465 1
   .0206026714295149 0
   .0277636740356684 0
   .0196803659200668 0
   .0267212316393852 0
   .0165427476167679 0
   .0143055869266391 0
   .0101226763799787 0
  .00400801096111536 0
 .000978711643256247 0
 -.00439134472981095 0
 -.00454534729942679 0
  -.0143248066306114 0
   -.011288495734334 0
   .0279180258512497 0
 -.00377739616669714 0
  -.0409102328121662 0
  -.0474172681570053 0
  -.0310680456459522 0
  -.0386427640914917 0
  -.0256036352366209 0
  -.0288705918937922 0
  -.0165830813348293 0
  -.0249304473400116 0
 -.00053030822891742 0
  .00246931938454509 0
  .00948905479162931 0
  .00401286501437426 0
  .00637947721406817 0
   .0151412636041641 0
 -.00107350014150143 0
   .0195041596889496 0
   .0210727192461491 0
   .0152838360518217 1
   .0308888535946608 0
    .013561112806201 0
   .0336562767624855 0
   .0239443592727184 0
   .0416115894913673 0
   .0360375605523586 0
   .0322027541697025 0
    .034945972263813 0
   .0158937368541956 0
   .0229483339935541 0
   .0182781536132097 0
   .0197849161922932 0
   .0217392519116402 0
 -.00598309794440866 1
-.000160088049597107 0
  .00276818196289241 0
 -.00900282058864832 0
   .0116364993155003 0
  -.0123568596318364 0
  -.0567549951374531 0
  -.0712292939424515 0
  -.0627420172095299 0
  -.0416260808706284 0
  -.0252468157559633 0
  -.0374706014990807 0
 -.00310045573860407 0
   .0114531088620424 0
   .0144753698259592 0
   .0150865437462926 0
     .03155517578125 0
   .0127113899216056 0
   .0142475571483374 0
   .0083512905985117 0
    .018443463370204 0
  .00539986090734601 0
  .00647167721763253 0
  .00456631183624268 0
   .0124574471265078 0
   .0133043806999922 0
  .00886440742760897 0
   .0223146472126246 0
   .0159194134175777 0
   .0109417755156755 0
end

I hope I explained this clearly, and thank you for your help.

[Clyde Schechter]

Well, my best guess is that you have misunderstood the paper. I think that it is far more likely that they used the difference in logarithms than the logarithm of the difference. Even the notation in the equation you show suggests that. Assuming that l represents logarithm, D.lpsemp would be the difference in logarithms of psemp, not the log of the difference. This would also be quite sensible since the difference in logarithms would be equivalent to the logarithm of the ratio, i.e. the logarithm of the growth ratio.

If they really did use the logarithm of the difference, then either they had some unusual data in which these differences were always positive, or they did not accurately describe their methods: any adequate description would have had to include an explanation of how they handled negative differences. I suggest you re-read the methods section in the article carefully. Perhaps there is some other explanation buried in there. But I assure you that they cannot possibly have used the logarithm of differences in data where some of the differences are negative. It is a mathematical impossibility.

[Clyde Schechter]

So, I think your rent growth variable is in the form of a relative change. If you want to work with the difference of the log of rent, which is the same as the log of the ratio, then we first have to convert that to a ratio. If rent growth is .0024519, then the rent growth ratio is 1.002451. More generally, the growth ratio will be 1 + growth. Then you can take the log of that.

gen log_ratio = log(1+RentGrowth)

This log_ratio variable is the same as the difference in log of rent. I'm pretty sure that's what is needed.

Added: By the way, because all these RentGrowth numbers are pretty close to zero, log_ratio will only be minimally different from RentGrowth itself.

As to whether it will significantly alter your results, you have not said anything about what your results are so far, nor how you arrived at them, so nobody can say. But just do it and see. In any case, you should not pick your analysis based on what results it gives. You should pick the analysis that is most credible in its own right and then let the proverbial chips fall where they may.

Finally, please don't use screenshots to show data. Had it been necessary to bring the data into Stata to test out code, there is no way to import data from a screenshot. Screenshots are the worst possible way to show data. The best way is -dataex-. And you know how to use it, since you used it in your original post. Stick with -dataex-: it gives a complete and faithful replica of your Stata example, it's easy for people to work with. And, as you must have noticed, its easier and quicker on your end then posting a screenshot, too.

[Clyde Schechter]

s there a way for me to show my outputs without posting a screenshot?

Yes. Just copy them from the Results window or your log file to your clipboard and then paste them into the forum editor between code delimiters. (If you don't know how to do code delimiters, read Forum FAQ #12.) The result will look a lot like a -dataex- output and will be easily readable.

As to the other questions you pose, none of them are, as far as I can see, about Stata or statistics. They are substantive questions about how to model your particular problem. So you need advice from an expert in economics or some closely related field. I have no expertise in those areas and can't help you. There are others on this Forum who specialize in those areas and I hope one of them will pick up the thread from here.