General pointers on writing an empirical social science paper€¦ · General pointers on writing...
Transcript of General pointers on writing an empirical social science paper€¦ · General pointers on writing...
![Page 1: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/1.jpg)
General pointers on writing an empirical social
science paper
–Organize ideas before writing
–Define important terms and use terms consistently
–State hypothesis/goal/research question
–Describe methods
– Interpret results (in addition to stating them)
–Avoid excess verbiage
–Have others read and critique
–Edit, edit, edit
![Page 2: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/2.jpg)
General pointers on writing an empirical social
science paper
–Organize ideas before writing
–Define important terms and use terms consistently
–State hypothesis/goal/research question
–Describe methods
– Interpret results (in addition to stating them)
–Avoid excess verbiage
–Have others read and critique
–Edit, edit, edit
![Page 3: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/3.jpg)
Will go through each of these sections in turn and tell
you what elements each should contain. (Which
sections you actually write will depend on your paper.)
– Introduction
–Data section
–Methods and Results
–(Conclusion)
![Page 4: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/4.jpg)
Will illustrate elements of each of these sections with examples drawn from four papers (that I happen to know well):
– “Diversity, Social Goods Provision, and Performance in the Firm,” JEMS 2014
– “Strategic Entry Deterrence and the Behavior of Pharmaceutical Incumbents Prior to Patent Expiration,” AEJMicro, 2011
– “Countervailing Power in Wholesale Pharmaceuticals,” JIE, 2010
– “Search, Obfuscation, and Price Elasticities on the Internet,” EMA, 2009
![Page 5: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/5.jpg)
Introduction
• Motivate why topic interesting/important
• Clearly state research question and describe (in non-
technical terms) how you will analyze
• Describe what others have done and how what you’re
doing fits in
• Foreshadow results
![Page 6: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/6.jpg)
Introduction
• Motivate why topic interesting/important
• Clearly state research question and describe (in non-
technical terms) how you will analyze
• Describe what others have done and how what you’re
doing fits in
• Foreshadow results
focus on this one
![Page 7: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/7.jpg)
Introduction
• Motivate why topic interesting/important
![Page 8: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/8.jpg)
Introduction
• Motivate why topic interesting/important
![Page 9: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/9.jpg)
Introduction
• Clearly state research question and describe (in non-
technical terms) how you will analyze
….
![Page 10: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/10.jpg)
Introduction
• Clearly state research question and describe (in non-
technical terms) how you will analyze
![Page 11: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/11.jpg)
Introduction
• Clearly state research question and describe (in non-
technical terms) how you will analyze partial answer
![Page 12: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/12.jpg)
Introduction
• Describe what others have done and how what you’re
doing fits in
![Page 13: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/13.jpg)
Introduction
• Foreshadow results
![Page 14: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/14.jpg)
Introduction
• Foreshadow results
![Page 15: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/15.jpg)
Data
• Cite sources for data
• Describe any special data treatments you’ve
performed
• Describe structure of data set (# obs, level of
observation, period of time covered, etc.)
• Present variable definitions
• Present summary statistics
• Discuss interesting facts, observations, shortcomings
Probably need tables
![Page 16: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/16.jpg)
Data
• Cite sources for data
![Page 17: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/17.jpg)
Data
• Describe any special data treatments you’ve
performed
![Page 18: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/18.jpg)
Data
• Describe structure of data set (# obs, level of
observation, period of time covered, etc.)
Leave clues in the text---other crucial information will be contained in the tables.
![Page 19: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/19.jpg)
Data
• Present variable definitions
• Present summary statisticsProbably need tables
Pretty self-explanatory namesReport all of these
![Page 20: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/20.jpg)
Note that there’s a lot of information in these two tables about structure of the
data set
Also note that these are not computer output!!!
![Page 21: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/21.jpg)
Note that there’s a lot of information in these two tables about structure of the
data set
And they do not include sci notation and a million significant digits.
![Page 22: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/22.jpg)
Data---do this for the data-driven version
• Discuss interesting facts, observations, shortcomings.
– Show us correlation tables
– Plot histograms of variables
–Create XY plots of pairs of variables
–Test whether two sets of observations are from the
same distribution
– Exhibit the truncation or top-coding of a variable
– Be creative, but selective---include what is interesting or
surprising or important and tell us why
![Page 23: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/23.jpg)
Data---do this for the data-driven version
• Discuss interesting facts, observations, shortcomings.
– Show us correlation tables
– Plot histograms of variables
–Create XY plots of pairs of variables
–Test whether two sets of observations are from the
same distribution
– Exhibit the truncation or top-coding of a variable
– Be creative, but selective---include what is interesting or
surprising or important and tell us why
Don’t forget this part!!
![Page 24: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/24.jpg)
Methods and Results
• Reiterate objective of paper
• Describe method
• Present results in table
• Describe and interpret results in text
![Page 25: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/25.jpg)
Methods and Results
• Reiterate objective of paper
Note how this language is drawing a clear connection between the paper’s objective
and the regressions we are about to describe.
A good results section should draw that connection repeatedly.
![Page 26: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/26.jpg)
Methods and Results
• Describe method
![Page 27: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/27.jpg)
Methods and Results
• Present results in table
Different specifications arranged in columns
Significance indicated by both SEs
and asterisks
No scientific notation!!!
![Page 28: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/28.jpg)
Methods and Results
• Describe and interpret results in text
![Page 29: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/29.jpg)
Methods and Results
• Describe and interpret results in text
![Page 30: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/30.jpg)
Methods and Results
• Describe and interpret results in text
Again, this language is drawing a clear connection between the paper’s objective
and the regression results---it is not enough to just repeat the numbers from the
table.
![Page 31: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/31.jpg)
Conclusion---probably not necessary for you
• Varies greatly by writer. If you do include one, it
should be viewed as an opportunity to make your
final case---this is the last thing that the reader
reads.
–Reiterate results
–Discuss policy implications
–Discuss future research
–Tie loose ends together
![Page 32: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/32.jpg)
Additional comments
• Structure and style of papers can and do deviate
from what I have presented. If you have good
reasons, deviating is fine, but essential elements
should not be omitted.
• My example papers were 20-35 pages long---yours
will be shorter.
• Clarity is always our primary goal---nothing else can
be achieved without clarity.
![Page 33: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/33.jpg)
Lecture 16: Running regressions–Practical issues
Prof. Esther Duflo
14.310x
1 / 20
![Page 34: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/34.jpg)
Practical issues with regression
• Reading a regression output
• Dummy Variables
• Other Functional Form issues
2 / 20
![Page 35: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/35.jpg)
Reading a regression output in R
3 / 20
![Page 36: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/36.jpg)
What R gives you when you typesummary(...)
• Intercept
• Coe�cients and standard errors of estimated coe�cients
• The T statistics: testing the hypothesis that each coe�cientis zero.
• And the associated p value, and even stars if you cannot readp value!
• The R squared
• The F stat of the regression: test the hypothesis that allcoe�cients are zero
4 / 20
![Page 37: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/37.jpg)
Doing more with the model
• The coe�cients are stored in a vector called ”coef”
• The variance covariance matrix of the coe�cient is a matrixcalled ”vcov”
• You can use the ”hypothesistesting” in the library ”car” totest any linear hypothesis of the form R� = 0 as seen in lastlecture
• you can export the coe�cients and the standard errors in adata frame (or use the ”stargazer” package to do much betterthan that... we will see it later hopefully)
• you can visualize prediction and residuals (see R code)
5 / 20
![Page 38: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/38.jpg)
Dummy Variables
Yi = ↵+ �Di + ✏i
Di is a dummy variable , or an indicator variable, if it takes thevalue 1 if the observation is in group A, and 0 if in group B.Example:
• RCT: 1 if in treatment group , 0 otherwise
• 1 if male, 0 if female
• 1 before great depression, 0 after
• 1 before generic substitution act passed, 0 otherwise,
• 1 if the house has a deck in the backyard, 0 otherwise,
6 / 20
![Page 39: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/39.jpg)
Interpretation
Yi = ↵+ �Di + ✏i
Without any control variable, it is easy to verify that b� = YA �YB .So you can always estimate the di↵erence between treatment andcontrol group for an RCT using an OLS regression framework.
7 / 20
![Page 40: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/40.jpg)
From a categorical variable to dummyvariables
• What if you don’t have two groups, but, say, 50 (e.g. 50 states):Your original variable is takes discrete values 1 to 50.
• It usually does not make much sense to include it directly as aregressor
• Transform it into 50 dummy variables: for each state, the dummy= 1 if the observation is from that state, and 0 otherwise.
• Careful, what happens if you introduce all of them and theconstant?
• R will complain about multi-colinearity.• So what do we do?• We typically omit ONE group (if we don’t do it, R may do it forus), and then what is the interpretation of each coe�cient?
• It is the di↵erence between the value of this group and the valuefor the omitted (reference) group.
8 / 20
![Page 41: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/41.jpg)
From a categorical variable to dummyvariables
• What if you don’t have two groups, but, say, 50 (e.g. 50 states):Your original variable is takes discrete values 1 to 50.
• It usually does not make much sense to include it directly as aregressor
• Transform it into 50 dummy variables: for each state, the dummy= 1 if the observation is from that state, and 0 otherwise.
• Careful, what happens if you introduce all of them and theconstant?
• R will complain about multi-colinearity.• So what do we do?
• We typically omit ONE group (if we don’t do it, R may do it forus), and then what is the interpretation of each coe�cient?
• It is the di↵erence between the value of this group and the valuefor the omitted (reference) group.
8 / 20
![Page 42: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/42.jpg)
From a categorical variable to dummyvariables
• What if you don’t have two groups, but, say, 50 (e.g. 50 states):Your original variable is takes discrete values 1 to 50.
• It usually does not make much sense to include it directly as aregressor
• Transform it into 50 dummy variables: for each state, the dummy= 1 if the observation is from that state, and 0 otherwise.
• Careful, what happens if you introduce all of them and theconstant?
• R will complain about multi-colinearity.• So what do we do?• We typically omit ONE group (if we don’t do it, R may do it forus), and then what is the interpretation of each coe�cient?
• It is the di↵erence between the value of this group and the valuefor the omitted (reference) group.
8 / 20
![Page 43: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/43.jpg)
From a categorical variable to dummyvariables
• What if you don’t have two groups, but, say, 50 (e.g. 50 states):Your original variable is takes discrete values 1 to 50.
• It usually does not make much sense to include it directly as aregressor
• Transform it into 50 dummy variables: for each state, the dummy= 1 if the observation is from that state, and 0 otherwise.
• Careful, what happens if you introduce all of them and theconstant?
• R will complain about multi-colinearity.• So what do we do?• We typically omit ONE group (if we don’t do it, R may do it forus), and then what is the interpretation of each coe�cient?
• It is the di↵erence between the value of this group and the valuefor the omitted (reference) group.
8 / 20
![Page 44: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/44.jpg)
with other variables in the regression
With other variables in the regression
Yi = ↵+ �Di + Xi� + ✏i
In that case � is the di↵erence in intercept between group A andgroup B. This is the most frequent way that RCT are analyzed:the matrix X are “control” variables: things that did not a↵ect theassignment but may have been di↵erent at baseline.
9 / 20
![Page 45: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/45.jpg)
Dummy variables and InteractionsNow imagine you have two sets of dummy variables, say,Treatment and control, and Male and Female.You can run:
Yi = ↵+ �Di + �Mi + �Mi ⇤ Di + ✏i
How do we interpret these coe�cients:
• b↵: An estimate of mean for women in the control group• b�: An estimate of the di↵erence between the treatment andcontrol group means for women [we call this the treatmentmain e↵ect]
• b�: An estimate of the di↵erence between Males and Females.[we call this the gender main e↵ect]
• b� An estimate of the di↵erence between the treatment e↵ectfor males and for female. [we call this the interaction e↵ect]
How do you obtain, for example, an estimate of the mean formales?How do you obtain an estimate of the treatment e↵ect for males?
10 / 20
![Page 46: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/46.jpg)
Dummy variables and InteractionsNow imagine you have two sets of dummy variables, say,Treatment and control, and Male and Female.You can run:
Yi = ↵+ �Di + �Mi + �Mi ⇤ Di + ✏i
How do we interpret these coe�cients:• b↵:
An estimate of mean for women in the control group• b�: An estimate of the di↵erence between the treatment andcontrol group means for women [we call this the treatmentmain e↵ect]
• b�: An estimate of the di↵erence between Males and Females.[we call this the gender main e↵ect]
• b� An estimate of the di↵erence between the treatment e↵ectfor males and for female. [we call this the interaction e↵ect]
How do you obtain, for example, an estimate of the mean formales?How do you obtain an estimate of the treatment e↵ect for males?
10 / 20
![Page 47: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/47.jpg)
Dummy variables and InteractionsNow imagine you have two sets of dummy variables, say,Treatment and control, and Male and Female.You can run:
Yi = ↵+ �Di + �Mi + �Mi ⇤ Di + ✏i
How do we interpret these coe�cients:• b↵: An estimate of mean for women in the control group
• b�: An estimate of the di↵erence between the treatment andcontrol group means for women [we call this the treatmentmain e↵ect]
• b�: An estimate of the di↵erence between Males and Females.[we call this the gender main e↵ect]
• b� An estimate of the di↵erence between the treatment e↵ectfor males and for female. [we call this the interaction e↵ect]
How do you obtain, for example, an estimate of the mean formales?How do you obtain an estimate of the treatment e↵ect for males?
10 / 20
![Page 48: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/48.jpg)
Dummy variables and InteractionsNow imagine you have two sets of dummy variables, say,Treatment and control, and Male and Female.You can run:
Yi = ↵+ �Di + �Mi + �Mi ⇤ Di + ✏i
How do we interpret these coe�cients:• b↵: An estimate of mean for women in the control group• b�:
An estimate of the di↵erence between the treatment andcontrol group means for women [we call this the treatmentmain e↵ect]
• b�: An estimate of the di↵erence between Males and Females.[we call this the gender main e↵ect]
• b� An estimate of the di↵erence between the treatment e↵ectfor males and for female. [we call this the interaction e↵ect]
How do you obtain, for example, an estimate of the mean formales?How do you obtain an estimate of the treatment e↵ect for males?
10 / 20
![Page 49: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/49.jpg)
Dummy variables and InteractionsNow imagine you have two sets of dummy variables, say,Treatment and control, and Male and Female.You can run:
Yi = ↵+ �Di + �Mi + �Mi ⇤ Di + ✏i
How do we interpret these coe�cients:• b↵: An estimate of mean for women in the control group• b�: An estimate of the di↵erence between the treatment andcontrol group means for women [we call this the treatmentmain e↵ect]
• b�: An estimate of the di↵erence between Males and Females.[we call this the gender main e↵ect]
• b� An estimate of the di↵erence between the treatment e↵ectfor males and for female. [we call this the interaction e↵ect]
How do you obtain, for example, an estimate of the mean formales?How do you obtain an estimate of the treatment e↵ect for males?
10 / 20
![Page 50: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/50.jpg)
Dummy variables and InteractionsNow imagine you have two sets of dummy variables, say,Treatment and control, and Male and Female.You can run:
Yi = ↵+ �Di + �Mi + �Mi ⇤ Di + ✏i
How do we interpret these coe�cients:• b↵: An estimate of mean for women in the control group• b�: An estimate of the di↵erence between the treatment andcontrol group means for women [we call this the treatmentmain e↵ect]
• b�:
An estimate of the di↵erence between Males and Females.[we call this the gender main e↵ect]
• b� An estimate of the di↵erence between the treatment e↵ectfor males and for female. [we call this the interaction e↵ect]
How do you obtain, for example, an estimate of the mean formales?How do you obtain an estimate of the treatment e↵ect for males?
10 / 20
![Page 51: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/51.jpg)
Dummy variables and InteractionsNow imagine you have two sets of dummy variables, say,Treatment and control, and Male and Female.You can run:
Yi = ↵+ �Di + �Mi + �Mi ⇤ Di + ✏i
How do we interpret these coe�cients:• b↵: An estimate of mean for women in the control group• b�: An estimate of the di↵erence between the treatment andcontrol group means for women [we call this the treatmentmain e↵ect]
• b�: An estimate of the di↵erence between Males and Females.[we call this the gender main e↵ect]
• b� An estimate of the di↵erence between the treatment e↵ectfor males and for female. [we call this the interaction e↵ect]
How do you obtain, for example, an estimate of the mean formales?How do you obtain an estimate of the treatment e↵ect for males?
10 / 20
![Page 52: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/52.jpg)
Dummy variables and InteractionsNow imagine you have two sets of dummy variables, say,Treatment and control, and Male and Female.You can run:
Yi = ↵+ �Di + �Mi + �Mi ⇤ Di + ✏i
How do we interpret these coe�cients:• b↵: An estimate of mean for women in the control group• b�: An estimate of the di↵erence between the treatment andcontrol group means for women [we call this the treatmentmain e↵ect]
• b�: An estimate of the di↵erence between Males and Females.[we call this the gender main e↵ect]
• b�
An estimate of the di↵erence between the treatment e↵ectfor males and for female. [we call this the interaction e↵ect]
How do you obtain, for example, an estimate of the mean formales?How do you obtain an estimate of the treatment e↵ect for males?
10 / 20
![Page 53: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/53.jpg)
Dummy variables and InteractionsNow imagine you have two sets of dummy variables, say,Treatment and control, and Male and Female.You can run:
Yi = ↵+ �Di + �Mi + �Mi ⇤ Di + ✏i
How do we interpret these coe�cients:• b↵: An estimate of mean for women in the control group• b�: An estimate of the di↵erence between the treatment andcontrol group means for women [we call this the treatmentmain e↵ect]
• b�: An estimate of the di↵erence between Males and Females.[we call this the gender main e↵ect]
• b� An estimate of the di↵erence between the treatment e↵ectfor males and for female. [we call this the interaction e↵ect]
How do you obtain, for example, an estimate of the mean formales?How do you obtain an estimate of the treatment e↵ect for males?
10 / 20
![Page 54: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/54.jpg)
Dummy variables and InteractionsNow imagine you have two sets of dummy variables, say,Treatment and control, and Male and Female.You can run:
Yi = ↵+ �Di + �Mi + �Mi ⇤ Di + ✏i
How do we interpret these coe�cients:• b↵: An estimate of mean for women in the control group• b�: An estimate of the di↵erence between the treatment andcontrol group means for women [we call this the treatmentmain e↵ect]
• b�: An estimate of the di↵erence between Males and Females.[we call this the gender main e↵ect]
• b� An estimate of the di↵erence between the treatment e↵ectfor males and for female. [we call this the interaction e↵ect]
How do you obtain, for example, an estimate of the mean formales?
How do you obtain an estimate of the treatment e↵ect for males?
10 / 20
![Page 55: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/55.jpg)
Dummy variables and InteractionsNow imagine you have two sets of dummy variables, say,Treatment and control, and Male and Female.You can run:
Yi = ↵+ �Di + �Mi + �Mi ⇤ Di + ✏i
How do we interpret these coe�cients:• b↵: An estimate of mean for women in the control group• b�: An estimate of the di↵erence between the treatment andcontrol group means for women [we call this the treatmentmain e↵ect]
• b�: An estimate of the di↵erence between Males and Females.[we call this the gender main e↵ect]
• b� An estimate of the di↵erence between the treatment e↵ectfor males and for female. [we call this the interaction e↵ect]
How do you obtain, for example, an estimate of the mean formales?How do you obtain an estimate of the treatment e↵ect for males?
10 / 20
![Page 56: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/56.jpg)
Interaction of a dummy variable and acontinuous variable
11 / 20
![Page 57: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/57.jpg)
Practical issues with regression
• Dummy Variables
• Other Functional Form issues
12 / 20
![Page 58: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/58.jpg)
Other functional form issues
• Transforming the dependent variable
• Non linear transformations of the independent variables
13 / 20
![Page 59: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/59.jpg)
Transformations of the dependent variable
• Suppose Yi = AX
�1
1i X�2
2i e✏i then run linear regression
log(Yi ) = �0
+ �1
logX
1i + �2
logX
12
+ ✏i
to estimate �1
and �2
. Note that �1
and �2
are elasticities:when X
1
changes by 1%, Y changes by �1
% .
• Returns to education formulation
logYi = �0
+ �1
Si + ✏i
When education increases by 1 year, wages increases by�1
⇥%.
14 / 20
![Page 60: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/60.jpg)
Transformations of the dependent variable
• Box Cox TransformationSuppose Yi =
1
�0
+�1
X1i+�
2
X2i+✏i
then run regression
1
Yi= �
0
+ �1
X
1i + �2
X
2i + ✏i
• Discrete choice modelSuppose
Pi =e
�0
+�1
X1i+�
2
X2i+✏i
1 + e
�0
+�1
X1i+�
2
X2i+✏i
Pi is the percentage of individuals choosing a particular option(e.g. buying a particular car)then run regression:
Yi = log(Pi
1� Pi) = �
0
+ �1
X
1i + �2
X
2i + ✏i
15 / 20
![Page 61: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/61.jpg)
Non linear transformation of theindependent variables
• When running a kernel regression as exploratory analysis wemay realize that the relationship between two variables doesnot appear to be linear.
• Does it mean we cannot run OLS?
• No!
• We can use polynomial or other transformations of the datato represent non linearities
• or partition the range of X .
16 / 20
![Page 62: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/62.jpg)
Non linear transformation of theindependent variables
• When running a kernel regression as exploratory analysis wemay realize that the relationship between two variables doesnot appear to be linear.
• Does it mean we cannot run OLS?
• No!
• We can use polynomial or other transformations of the datato represent non linearities
• or partition the range of X .
16 / 20
![Page 63: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/63.jpg)
Polynomial models
Yi = �0
+ �1
X
1i + �2
X
2
1i + · · ·+ �kXk1i + ✏i
• You can chose straight polynomial, or series expansion, ororthogonal polynomials or whatever.
• If you assume that the model is known, this is just standardOLS. You may want to plot the curve, compute the derivativewith respect to X at key points, etc.
• If you assume that the model is now known, this is anon-parametric method: you realize there is bias (because theshape is never quite perfect) and variance (as you add moreXs) and you promise to add more terms as the number ofobservation increases. This is called series regression.
17 / 20
![Page 64: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/64.jpg)
Other non linear transformations
• Take log of X
• Interact the X , such as the slope of one depends on the levelof another.
• Potentially lots of variables and their transformations... Howto chose? This is where machine learning tools can becomehandy (more on that later!)
18 / 20
![Page 65: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/65.jpg)
Using dummies for approximation
• Partition the range of X is interval, X 0, . . .X J
• Define the dummies as:D
1i = I
[X 0X1i<X 1
]
D
2i = I
[X 1X1i<X 2
]
. . .
• you can run regression:
Yi = �1
D
1i + �2
D
2i + · · ·+ �JDji + ✏i
(note no intercept. why?)
• Define Piece wise linear variables as:S
1i = I
[X 0X1i<X 1
]
(X1i � X
1) S2i = I
[X 1X1i<X 2
]
(X1i � X
2)
• Run regression
Yi = �1
X
1i + �2
S
1i + · · ·+ �JSj�1i + ✏i
19 / 20
![Page 66: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/66.jpg)
Using dummies for approximation
• Partition the range of X is interval, X 0, . . .X J
• Define the dummies as:D
1i = I
[X 0X1i<X 1
]
D
2i = I
[X 1X1i<X 2
]
. . .
• you can run regression:
Yi = �1
D
1i + �2
D
2i + · · ·+ �JDji + ✏i
(note no intercept. why?)
• Define Piece wise linear variables as:S
1i = I
[X 0X1i<X 1
]
(X1i � X
1) S2i = I
[X 1X1i<X 2
]
(X1i � X
2)
• Run regression
Yi = �1
X
1i + �2
S
1i + · · ·+ �JSj�1i + ✏i
19 / 20
![Page 67: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/67.jpg)
Using dummies for approximation
• Partition the range of X is interval, X 0, . . .X J
• Define the dummies as:D
1i = I
[X 0X1i<X 1
]
D
2i = I
[X 1X1i<X 2
]
. . .
• you can run regression:
Yi = �1
D
1i + �2
D
2i + · · ·+ �JDji + ✏i
(note no intercept. why?)
• Define Piece wise linear variables as:S
1i = I
[X 0X1i<X 1
]
(X1i � X
1) S2i = I
[X 1X1i<X 2
]
(X1i � X
2)
• Run regression
Yi = �1
X
1i + �2
S
1i + · · ·+ �JSj�1i + ✏i
19 / 20
![Page 68: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/68.jpg)
Using dummies for approximation
• Partition the range of X is interval, X 0, . . .X J
• Define the dummies as:D
1i = I
[X 0X1i<X 1
]
D
2i = I
[X 1X1i<X 2
]
. . .
• you can run regression:
Yi = �1
D
1i + �2
D
2i + · · ·+ �JDji + ✏i
(note no intercept. why?)
• Define Piece wise linear variables as:S
1i = I
[X 0X1i<X 1
]
(X1i � X
1) S2i = I
[X 1X1i<X 2
]
(X1i � X
2)
• Run regression
Yi = �1
X
1i + �2
S
1i + · · ·+ �JSj�1i + ✏i
19 / 20
![Page 69: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/69.jpg)
Locally Linear Regression• What size of interval should we chose?• This should by now sound very familiar: either you are willingto assume that you know the shape of the function: Then,just cut it as you know it is relevant.
• Or.... we are trying to guess the shape of the function• And then we have the familiar bias/variance trade o↵: we arenow in fact performing a non parametric regression techniqueknows a locally linear regression: around each point where weare interested in evaluating the function, we run a weightedregression of Yi on Xi , where the weights will be given by aKernel, for observations in a bandwidth. We take thepredicted value from the regression as best predictor for Yi .So it is exactly like a Kernel regression, but we use a linearregression in each little interval instead!
• Why on earth?
• It has better properties (especially at the boundaries)• And the slope is often of interest• This is what R do when you specify the option ”LOESS” (or
don’t specify anything, and your data set is not very large, inR).
20 / 20
![Page 70: General pointers on writing an empirical social science paper€¦ · General pointers on writing an empirical social science paper –Organize ideas before writing –Define important](https://reader033.fdocuments.us/reader033/viewer/2022050422/5f91b0bd37ba7b5f2901daf0/html5/thumbnails/70.jpg)
Locally Linear Regression• What size of interval should we chose?• This should by now sound very familiar: either you are willingto assume that you know the shape of the function: Then,just cut it as you know it is relevant.
• Or.... we are trying to guess the shape of the function• And then we have the familiar bias/variance trade o↵: we arenow in fact performing a non parametric regression techniqueknows a locally linear regression: around each point where weare interested in evaluating the function, we run a weightedregression of Yi on Xi , where the weights will be given by aKernel, for observations in a bandwidth. We take thepredicted value from the regression as best predictor for Yi .So it is exactly like a Kernel regression, but we use a linearregression in each little interval instead!
• Why on earth?• It has better properties (especially at the boundaries)• And the slope is often of interest• This is what R do when you specify the option ”LOESS” (or
don’t specify anything, and your data set is not very large, inR).
20 / 20