Use Dataset 1 to answer the questions.
Assignment 3
Suppose you want to examine whether the mobile commerce channel (e.g., Smartphone) usage encourages online shopping, i.e., the causal effect of mobile channel adoption on the purchase amount.
Q1. Tell me your conjecture before analysis. Would mobile channel adoption encourage online shopping? Why do you think so?
Q2. Load ¡°OnlineMember.dta.¡± Exclude business buyers. Also, exclude those who made the first purchase after 20090601 from the data. Likewise, load ¡°MobileMember.dta¡± and exclude such users from the data. Apply one-to-one matching without replacement based on ¡®Gender¡¯ and ¡®Age.¡¯ Make sure the adopters and non-adopters are well-matched regarding the two variables. Save the user IDs of matched adopters and matched non-adopters as a separate file named ¡°Counterpart.csv¡±.
Q3. Load ¡°MobileOrder.dta¡± and generate a dummy variable, Mobile, where 1= if Mall == “03” & (AccessRoute==”1000132495″ | AccessRoute ==”1000132496″ | AccessRoute ==”1000013091″) and 0=otherwise, which indicates whether the transaction happened through the mobile channel. Likewise, load ¡°OnlineOrder.dta¡± and make a dummy variable, Mobile, which should have the value of 0 for all observations. First six digits of ¡°BasketID¡± represent the month when the transaction was made. Generate a variable named “Month” for both ¡°MobileOrder.dta¡± and ¡°OnlineOrder.dta¡± by extracting a substring from ¡°BasketID.¡± Keep only four variables, “MemberID”, “OrderPrice”, “Mobile”, and “Month,” in both the adopter and non-adopter data tables for a next step.
Q4. Delete transactions of unmatched samples from the adopter and non-adopter data tables (Hint: Use the file saved in Q2). Generate two-year user-month panel data tables for adopters and non-adopters (from 200906 to 201105). Refer to the example below. For example, let us assume that a user numbered 1234 made four transactions in total, two in 201006, one in 201008, and one in 201011. It means that the user did not make any purchase in other months including 200906, 200907, … , 201005, 201007, 201009, 201010, 201012, …, 201105. The resulting user-month panel data table should look like Table B. ¡°M_OrderPrice¡± is the monthly OrderPrice. ¡°M_Mobile¡± has the value of 1 if the month contains at least one mobile transaction and 0 otherwise (Hint: You can easily make this variable by using ¡°Mobile¡±). In this panel data, each user should have 24 observations.
Table A. Original Order table
Table B. User-month panel table
MemberID
Month
OrderPrice
Mobile
1234
201006
100
0
1234
201006
200
0
1234
201008
200
1
1234
201011
300
0
2224
…
…
…
MemberID
Month
M_OrderPrice
M_Mobile
1234
200906
0
0
1234
200907
0
0
1234
200908
0
0
1234
200909
0
0
…..
1234
201006
300
0
1234
201007
0
0
1234
201008
200
1
1234
201009
0
0
1234
201010
0
0
1234
201011
300
0
1234
201012
0
0
…
2224
…
…
…
Q5. Generate the variable After, for adopters, which indicates the month after adoption. Conduct t-test to compare M_OrderPrice before and after the adoption by using After. Interpret the result. Does it imply the causal relationship? Why or why not?
Q6. Generate the variable After for non-adopters, which indicates the month after their counterpart¡¯s (matched adopter¡¯s) adoption. Make a single order dataset by appending adopter and non-adopter data tables. Regress M_OrderPrice on After, Adopt, and the interaction of both (After *Adopt). Interpret the result.
Q7. Are you convinced with the model specification above? Why or why not? How can we improve the model to examine the causal effect?