search

DS4PS / cpp-525-spr-2020

Course shell for CPP 525 Foundations of Program Evaluation III for Spring 2020.
http://ds4ps.org/cpp-525-spr-2020/
0 stars 0 forks source link

LAB-02 #3

Open sunaynagoel opened 4 years ago

sunaynagoel commented 4 years ago

Two questions- a. Is Week-02 lecture and lab open? when I click on WK-02 lecture it brings me back to Week 01. b. Are we going to have zoom session this friday? Respectfully ~Nina

lecy commented 4 years ago

The lab should be open.

Regarding a zoom session on Friday, I believe one is scheduled. I will let @itsapara confirm?

I know there was some confusion about times last night because the announcement was based on Chicago time and not Arizona time, so please note:

05:00 PM Arizona Time (07:00 PM Central Time) Every week on Mon, Fri, until April 10

Join from PC, Mac, Linux, iOS or Android: https://asu.zoom.us/j/511875479

sunaynagoel commented 4 years ago

The lab should be open.

Regarding a zoom session on Friday, I believe one is scheduled. I will let @itsapara confirm?

I know there was some confusion about times last night because the announcement was based on Chicago time and not Arizona time, so please note:

05:00 PM Arizona Time (07:00 PM Central Time) Every week on Mon, Fri, until April 10

  • Mar 23, 2020 07:00 PM
  • Mar 27, 2020 07:00 PM
  • Mar 30, 2020 07:00 PM
  • Apr 3, 2020 07:00 PM
  • Apr 6, 2020 07:00 PM
  • Apr 10, 2020 07:00 PM

Join from PC, Mac, Linux, iOS or Android: https://asu.zoom.us/j/511875479

Thank you so much.

itsapara commented 4 years ago

Hello The lab should be open. I had a meeting on Monday but noone attended. I will hold a second meeting on Friday and I hope to see you there.Please let me know if you cannot see the lab? On Wednesday, March 25, 2020, 12:07:38 PM CDT, sunaynagoel notifications@github.com wrote:

The lab should be open.

Regarding a zoom session on Friday, I believe one is scheduled. I will let @itsapara confirm?

I know there was some confusion about times last night because the announcement was based on Chicago time and not Arizona time, so please note:

05:00 PM Arizona Time (07:00 PM Central Time) Every week on Mon, Fri, until April 10

Join from PC, Mac, Linux, iOS or Android: https://asu.zoom.us/j/511875479

Thank you so much.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

sunaynagoel commented 4 years ago

Hello The lab should be open. I had a meeting on Monday but noone attended. I will hold a second meeting on Friday and I hope to see you there.Please let me know if you cannot see the lab? On Wednesday, March 25, 2020, 12:07:38 PM CDT, sunaynagoel notifications@github.com wrote: The lab should be open. Regarding a zoom session on Friday, I believe one is scheduled. I will let @itsapara confirm? I know there was some confusion about times last night because the announcement was based on Chicago time and not Arizona time, so please note: 05:00 PM Arizona Time (07:00 PM Central Time) Every week on Mon, Fri, until April 10 - Mar 23, 2020 07:00 PM - Mar 27, 2020 07:00 PM - Mar 30, 2020 07:00 PM - Apr 3, 2020 07:00 PM - Apr 6, 2020 07:00 PM - Apr 10, 2020 07:00 PM Join from PC, Mac, Linux, iOS or Android: https://asu.zoom.us/j/511875479 Thank you so much. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe.

I can see the lab now. Thank you. For zoom class, I think we got confused with the time. Next time I will keep the time zone in mind. Sorry about that. ~Nina

sunaynagoel commented 4 years ago

The data set shows a variable called "month". It contain two values - "0" or "10". I am not sure what is it representing. Thanks ~Nina

sunaynagoel commented 4 years ago

I do not understand Q4b. Q4b. How are b2 and b3 in this model different from b2 and b3 in the diff-in-diff model you estimated before?
Which model is being referred here as the one we have estimated before?

lecy commented 4 years ago

It looks like the data got corrupted. Give me one minute to fix that.

lecy commented 4 years ago

@sunaynagoel the dataset was overwritten by an old version of the code when i was consolidating some files. i was just re-creating the correct dataset and discovered a bug. i was running the model to make sure it makes sense and am getting some weird results. please give me a few minutes to update this - sorry for the bad data.

lecy commented 4 years ago

@sunaynagoel the data has been fixed and the question should be a lot more straight-forward now.

sunaynagoel commented 4 years ago

@sunaynagoel the data has been fixed and the question should be a lot more straight-forward now.

@lecy Thank you so much. Question 4 now makes more sense.

sunaynagoel commented 4 years ago

@lecy the Q4 part of the lab shows different number of observations for diff-in-diff in model and interrupted time series model. Diff-in-diff model has double the number of observations. Am I missing something?

lecy commented 4 years ago

@sunaynagoel

Note we only need the treatment group for this model since it is a reflexive design:

d2 <- filter( dat, group == "treat" )

If you scroll down to the DID example with a control group you will see the sample size is back to 1,248 since you include both groups in that model.

image

sunaynagoel commented 4 years ago

@sunaynagoel

Note we only need the treatment group for this model since it is a reflexive design:

d2 <- filter( dat, group == "treat" )

Thanks. I totally missed that part.

sunaynagoel commented 4 years ago

Hello @itsapara @lecy, There are two of us in the zoom class meeting right now. We were wondering if we are still meeting today? ~Nina

sunaynagoel commented 4 years ago

@lecy @itsapara How would you test the parallel lines assumption for the difference-in-difference model with this data? Write the code to create the appropriate data subset and model. Report your results.

For this question, should I create a new model or use the existing model. There would be nothing new to report on existing model.

~Nina

lecy commented 4 years ago

@sunaynagoel The parallel lines assumption is specifically to test whether the treatment and comparison groups experience the same pre-treatment trends. This is the identifying assumption of the model because it allows us to create the counterfactual, what we believe the treatment group mean would be if the treatment had not been administered: b0 + b1 + b2 (C1 + pre-treatment difference + trend).

Since we use C2-C1 to estimate trend for both groups, it has to be the case that the treatment group would experience the same trend independent of the treatment.

You would need to drop the post-treatment period and then using a DID model test whether the two lines run parallel. So using the large dots at 6 and 18 here approximately:

image

Does that make sense?

After the data steps the important part of the question is, which coefficient in the DID model would tell you in the lines are parallel?

castower commented 4 years ago

Hello @lecy @itsapara , I have a question concerning the Treatment*Post variable. Is this a calculation of Treatment multiplied by Post?

In some cases, that seems to be calculation:

But in others, like the 3rd row on this image, that doesn't seem to be the case (1x1 would equal 1). Am I overlooking something?

Thanks! Courtney

sunaynagoel commented 4 years ago

@sunaynagoel The parallel lines assumption is specifically to test whether the treatment and comparison groups experience the same pre-treatment trends. This is the identifying assumption of the model because it allows us to create the counterfactual, what we believe the treatment group mean would be if the treatment had not been administered: b0 + b1 + b2 (C1 + pre-treatment difference + trend).

Since we use C2-C1 to estimate trend for both groups, it has to be the case that the treatment group would experience the same trend independent of the treatment.

You would need to drop the post-treatment period and then using a DID model test whether the two lines run parallel. So using the large dots at 6 and 18 here approximately:

image

Does that make sense?

After the data steps the important part of the question is, which coefficient in the DID model would tell you in the lines are parallel?

Thanks, I understand it now.

lecy commented 4 years ago

@castower Correct, it should be treatment x post.

That was a typo on the image. It has been corrected.

image

castower commented 4 years ago

@lecy thank you!