LI Econometrics (08 29172)
Stata Assignment
In this assignment, you will explore correlates of earnings. You will use data on real individuals collected by the Office for National Statistics and published in the Quarterly Labour Force Survey.
In the course of working on this assignment, you will:
• be introduced to an important source of data for research on the UK (the UK Data Service);
• become familiar with processing and analysing data using Stata;
• interpret regression output;
• develop critical thinking about economic phenomena and econometric analysis of them.
Practical details
Word limit: no more than 1500 words. Reasonable use of tables and figures does not count toward the word limit.
Submission file: one file (in .pdf or .doc format).
Sections: Please label the sections and sub-sections clearly.
Figures and tables: should be numbered and titled appropriately. Tables should be formatted and presented as in standard economics journals (copy-pasted output from Stata is not acceptable).
Appendix: an appendix should be included at the end of the assignment, containing a copy of the Stata code used to obtain the results. This should be the exact code used to go from loading the dataset to generating the results presented (no more, no less). For convenience, you can simply copy and paste the .do file.
Preliminaries
You will have familiarised yourself with how to use the UK Data Service in the support session in week 6. Following the same steps, access and download “Quarterly Labour Force Survey, April-June, 2018 (SN 8381)".
0. Loading data and defining the sample
a) Load the main dataset (lfsp_aj18_eul.dta) in Stata.
b) We will focus on individuals reporting positive gross weekly earnings and not currently working towards a qualification. To keep only these observations, do:
keep if GRSSWK > 0 & QULNOW == 2
c) Check: the resulting dataset should contain 9141 observations. If this is not the case, something has gone wrong somewhere.
Section A
1. Region and earnings
a) Plot the distribution of weekly earnings (GRSSWK) in a histogram and provide a brief comment.
b) Now let us think about the region in which the respondents live. With reference to economic theory and intuition, why might there be regional differences in earnings?
c) Construct a new variable by taking the logarithm of gross weekly earnings, i.e. log(GRSSWK). Then, using the variable AGE, create a new variable that represents the squared age. Further, using the variable COUNTRY, that represents the country within the UK that the respondent lives in, create dummy variables for each one, i.e. England, Scotland, Wales and Northern Ireland (make sure that your Scotland variable includes both the categories of “Scotland” and “Scotland North of Caledonian Canal”). Using Wales as the baseline country, estimate the regression:
logGRSSWKi=α+γ1Agei+γ2Agei2+β1Englandi+β2Scotlandi+β3Northern Irelandi+ϵi (1)
Report your findings in a table, discuss the coefficients and comment on the country differences that you have found.
d) Using Stata’s test command, perform. three tests to show whether there are any statistical differences in earnings between residents living in (i) England and Scotland; (ii) England and Northern Ireland; and (iii) Scotland and Northern Ireland. Specifically for test (i), show also how it can be done by hand, i.e. by calculating the F statistic yourself (make sure you show all your calculations as well as the restricted model that you need to estimate in order to calculate the test statistic).
e) Now let’s consider only residents that live in England, by removing anyone that lives in the other nations (you can use the keep command for this). The resulting dataset should contain 7616 observations. Using the URESMC variable, create dummy variables for each region in England and estimate a regression that includes the age variables and these region dummies (use Merseyside as the base category as it has the fewest observations). Report your findings in a new table and make sure that your regional dummies have meaningful names. Are there significant regional differences in earnings? Do your findings support your economic intuition about regional effects, for example, do your results show a London effect?
2. Education
a) The variable HIQUL15D details the highest qualification that each respondent has achieved. Tabulate HIQUL15D to see what categories exist. Remove anyone that either did not answer or did not know their highest qualification, i.e. keep only those people that have a qualification of some sort or have no qualifications at all. This should leave you with 7521 observations. Create a dummy variable that takes the value 1 if someone has a degree or equivalent and 0 otherwise. Run the same regression as in part 1e) but now include your degree dummy as well. Report your findings in the same table from part 1e). With reference to links between education and region, how might differences between these results and those from part 1e) be explained?
b) Run a test to see whether all the region coefficients are the same. Then run the test again but exclude Inner London, Outer London and the rest of the southeast. Are these results what you expected?
Section B
3. Other factors
a) Pick an additional dimension on which information is available in the dataset (consult the user manual and codebook (lfs_user_guide_vol3_variabledetails2018.pdf) for details) and construct variable(s) which allow you to explore the relationship between this dimension and earnings. Without running additional regressions (yet), why does economic theory tell you to expect this dimension to matter for earnings? Do you think your chosen dimension might vary across English regions, or between respondents with and without a degree?
b) In the sample you used for part 2, estimate additional regressions to test whether your theoretical predictions hold in the data. Present these in a second table and discuss. What do your findings tell you about how your chosen dimension matters for earnings?
4. Taking a step back
a) In defining our sample, we kept only individuals reporting positive gross weekly earnings. Why might that affect any conclusions we draw about regional differences in labour market outcomes?
b) When conducting our regional analysis above, we considered respondents’ region of residence. The people residing in a region will be a mix of those who were born there (“non-movers”) and those who moved there from other regions or from abroad (“movers”). Suppose we could distinguish between “movers” and “non-movers”. Speculate on how you think the various results you have found in your analysis might differ between the two groups.