LAB -04

[sunaynagoel]

I have a few questions:

1. For question 2, the naive model will have race, education, age, and marital status as control variables?

naive.model <- lm ( data=data, HealthStatus ~ PublicHousing + Age + Race + Education + MaritalStatus )  

2. For question 2a, which is - Provide the full model table with stargazer. I am supposed to run a model which includes all the variable including the variables that can be Instrumental Variables ?

full.model <- lm ( data=data, HealthStatus ~ PublicHousing + Supply + WaitingTime + HealthBehavior + Stamp + ParentsHealthStatus+ Age + Race + Education + MaritalStatus )

3. For question 2b, What is the estimated effect of public housing? Is that statistically significant? How much does an additional month of public housing assistance decrease or increase health status?

Do I answer these questions based on the full model or the naive model?

Thank you

~Nina

[lecy]

I changed the phrasing a bit to make the instructions clear. Q2 is only looking at the "naive" model that does not account for our primary omitted variable of interest, healthy behaviors.

---

Q2

Let’s examine the estimated impact of the policy variable in a model that does not account for omitted variable bias.

Run a naive model that predict the effect of public housing assistance on health status. Include race, education, age, and marital status as control variables. We are not accounting for our omitted variable, healthy lifestyle, in this initial model.

Q2a: Report the results in a regression table with stargazer. (5 points)
Q2b: What is the estimated effect of public housing in this model? Is that statistically significant? How much does an additional month of public housing assistance decrease or increase health status? (5 points)

[sunaynagoel]

I changed the phrasing a bit to make the instructions clear. Q2 is only looking at the "naive" model that does not account for our primary omitted variable of interest, healthy behaviors.

Q2

Let’s examine the estimated impact of the policy variable in a model that does not account for omitted variable bias.

Run a naive model that predict the effect of public housing assistance on health status. Include race, education, age, and marital status as control variables. We are not accounting for our omitted variable, healthy lifestyle, in this initial model.

Q2a: Report the results in a regression table with stargazer. (5 points)
Q2b: What is the estimated effect of public housing in this model? Is that statistically significant? How much does an additional month of public housing assistance decrease or increase health status? (5 points)

Thank you

[castower]

@lecy
I have a question concerning Part 4 and Part 5 of the assignment. I only have one variable that meets all 3 criteria with .01 as the significance level for the correlation plot, should I lower this to .05 as the sig level?

[castower]

@lecy one more question:
Is the data dictionary for the lab correct that time in public housing is displayed in years and not months like the other variables? Specifically for the following statistics:

> summary(data$publichousing) %>% pander()

--------------------------------------------------
 Min.   1st Qu.   Median   Mean    3rd Qu.   Max. 
------ --------- -------- ------- --------- ------
  0      23.67    29.83    29.37    35.22     60  
--------------------------------------------------

The mean is 29 years, not 29 months, correct?

[lecy]

@castower There is a difference between a variable meeting the criteria for an IV, and it working well to remove bias. Specifically see the weak instruments problem. Q4 is asking about variables that meet the criteria.

[lecy]

@castower I did not create this data so I would need to double-check but I think that it is months, not years. Average stay in public housing is a little over 4 years.

https://www.huduser.gov/portal/sites/default/files/pdf/LengthofStay.pdf

Plus 60 years in public housing seems unrealistic!

I'll update the data dictionary to make that clear.

[castower]

@lecy thank you!

That makes sense for the variables and I thought the time periods seemed rather long, but I wanted to double-check.