1 Conflict and income in the long run (30 points)

We now link historical conflict data (HCED) with modern economic indicators using the PRIO-Grid. The goal is to assess how historical battles correlate with modern income across grid cells.

Spatial join (10 points)

Load required libraries and the cleaned HCED and PRIO-Grid data from before.
Make sure your HCED data is an sf object and transform it to the same CRS as PRIO-Grid, then perform the spatial join
Match each battle to a PRIO-Grid cell.¹
Aggregate the HCED events to a grid-level dataset, including the Number of battles overall and the number of war years, counting the years from 1500CE with at least one recorded battle.²
Merge the result back to the full grid and make it again an sf object with appropriate projection for plotting.
Set the number of battles and war years to zero for grid cells with no recorded battles.
Restrict the data to populated areas

Describe the data (5 points)

Make a table with descriptive statistics (Mean, SD, Min, Max, N) for the key variables in the merged grid (all cells).
In particular, show statistics for war years and number of battles, log night lights and log gross cell product, as well as population.
Briefly describe the data and comment on missing information
Geographic distribution of War years: Focusing again on the continent/region from before, make a map of the distribution of war years
Comment on the patterns you see, is there a visible association with the nightlight and population data?

Empirical approach (5 points)

To assess the empirical relationship between historical conflict and modern income, we will estimate 4 different models (simple OLS linear regression), with $\log(\textit{night lights})$ as dependent variable and war years on the right-hand side:

Without additional control variables
Controlling for selected geographic factors (e.g., mountains, land area, latitude, …)
Additionally controlling for population
Restricted to your previously chosen continent, controlling only for geographic factors

Write down the 4 equations and comment on the approaches. In particular, explain why it might be a good or bad idea to control for (a) geographic factors and (b) modern population if we want to assess the effect of historical conflict on income in the year 2000.

Estimation results (5 points)

Provide a combined table of the results, reporting the estimated coefficients with standard errors in parentheses. Do not show stars for significance levels.
Add a plot with the estimated coefficients for war years with 95% confidence intervals.
Briefly but precisely describe the results - are the coefficients significantly different from zero, are they larger than what you would consider practically equivalent?
Describe the differences you observe across the four models.

Discussion (5 points)

Discuss what we can learn from this exercise. In particular:

To what extend does the estimated coefficient represent the causal effect of historical conflict on modern income?
How the results differ from published findings?

Footnotes

Note that the data also contains information on larger battles, potentially stretching over multiple grid-cells. For the purpose of this assignment, please feel free to ignore this and use only the point coordinates.↩︎
HINT: First make a panel, where you group the data by grid cell and start year and indicate with a zero/one indicator if a battle was recorded, then group by grid id and count years with battles. Note that some battles last over multiple years - again feel free to ignore this for now (bonus points if you manage to handle this correctly).↩︎

# Conflict and income in the long run (**30 points**) We now link historical conflict data (HCED) with modern economic indicators using the PRIO-Grid. The goal is to assess how historical battles correlate with modern income across grid cells. ## Spatial join (10 points) - Load required libraries and the cleaned HCED and PRIO-Grid data from before. - Make sure your HCED data is an sf object and transform it to the same CRS as PRIO-Grid, then perform the spatial join - Match each battle to a PRIO-Grid cell.^[Note that the data also contains information on larger battles, potentially stretching over multiple grid-cells. For the purpose of this assignment, please feel free to ignore this and use only the point coordinates.] - Aggregate the HCED events to a grid-level dataset, including the Number of battles overall and the number of *war years*, counting the years from 1500CE with at least one recorded battle.^[HINT: First make a panel, where you group the data by grid cell and start year and indicate with a zero/one indicator if a battle was recorded, then group by grid id and count years with battles. Note that some battles last over multiple years - again feel free to ignore this for now (bonus points if you manage to handle this correctly).] - Merge the result back to the full grid and make it again an sf object with appropriate projection for plotting. - Set the number of battles and war years to zero for grid cells with no recorded battles. - Restrict the data to populated areas ## Describe the data (5 points) - Make a table with descriptive statistics (Mean, SD, Min, Max, N) for the key variables in the merged grid (all cells). - In particular, show statistics for war years and number of battles, log night lights and log gross cell product, as well as population. - Briefly describe the data and comment on missing information - Geographic distribution of *War years*: Focusing again on the continent/region from before, make a map of the distribution of war years - Comment on the patterns you see, is there a visible association with the nightlight and population data? ## Empirical approach (5 points) To assess the empirical relationship between historical conflict and modern income, we will estimate 4 different models (simple OLS linear regression), with $\log(\textit{night lights})$ as dependent variable and *war years* on the right-hand side: 1. Without additional control variables 2. Controlling for selected geographic factors (e.g., mountains, land area, latitude, ...) 3. Additionally controlling for population 4. Restricted to your previously chosen continent, controlling only for geographic factors Write down the 4 equations and comment on the approaches. In particular, explain why it might be a good or bad idea to control for (a) geographic factors and (b) modern population if we want to assess the effect of historical conflict on income in the year 2000. ## Estimation results (5 points) - Provide a combined table of the results, reporting the estimated coefficients with standard errors in parentheses. Do not show stars for significance levels. - Add a plot with the estimated coefficients for war years with 95% confidence intervals. - Briefly but precisely describe the results - are the coefficients significantly different from zero, are they larger than what you would consider practically equivalent? - Describe the differences you observe across the four models. ## Discussion (5 points) Discuss what we can learn from this exercise. In particular: - To what extend does the estimated coefficient represent the causal effect of historical conflict on modern income? - How the results differ from published findings?