For one Indian state of Rajasthan, 57000 postings for around 3700 unique bureaucratic jobs spanning several decades but most density is covered during the period 1998-2023. It is an unbalanced panel, most jobs have relatively recent first posting. This data is at district level. This data also contains the department and designation of any particular posting.
For one Indian state of Rajasthan, detailed election data consisting of election outcome and candidate characteristics, which includes assets, education and number of criminal charges against them. This data is at constituency level ( sub-district). One district has several assembly constituencies, and (almost)no constituency crosses district border.
Variable Construction
In the jobs posting data, I count the number of new "posting orders" are generated at district-year level. This is the outcome variable. Elections are every 5 years, and I can sum over a period of election cycle.
In the election data, I aggregate the assembly level data to district level. For each district-electionyear, I compute the raw sum of assets of the winner in that election. This gives me the total assets of all the winning candidates belonging to that district.
Figures
1. Number of postings orders always peak in the year after election.
2. Candidates with more assets are significantly correlated with higher churn at the bureaucracy
Looking into total churn over 5 years and after taking out district level fixed effects, ie deriving from within district variation in asset levels. One SD increase in assets (at district level) are correlated with 21% increase in bureaucratic churn (compared to mean). The regression coefficient has a t-stat of 3.8
We can also look at year specific effects. For example just looking at contemporaneous year effects, we find similar 18% effect size.
However, effects for all other years are noisy zero.
This is okay as individual year effects can be noisy and its okay for the effect to play out over the entire term limit.
3. There is some evidence of non-linearity in the above relationship.
Using the same specification, but adding a higher order degree 2 asset term in the regression, we get some non-linearity. However, the nonlinearity only kicks in far right part of the support and possibly not that interesting.
4. Number of criminal charges are not related with the bureaucratic churn. Number of criminal charges are also quite orthogonal to the asset levels.
Current/Next Steps
More dimensions can be analyzed here ( incumbent/alignment with the state party, etc). But I am currently focussing on moving one step ahead and compiling data about projects that brought big investments with them.
What I have in mind is that assuming big investment projects are decided at higher level than local politician, but local politician can still respond. A big investment project might attract a certain kind of politician because s/he expects that it provides them with more opportunities for holdups. This increases their interference with the bureaucracy and kicks in the phenomenon of interest to me.
I am looking into a few potential candidates like list of big projects, construction price index, etc that are likely to be correlated with local economic expectation.
Data Environment
Variable Construction
Figures
1. Number of postings orders always peak in the year after election.
2. Candidates with more assets are significantly correlated with higher churn at the bureaucracy
Looking into total churn over 5 years and after taking out district level fixed effects, ie deriving from within district variation in asset levels. One SD increase in assets (at district level) are correlated with 21% increase in bureaucratic churn (compared to mean). The regression coefficient has a t-stat of 3.8
We can also look at year specific effects. For example just looking at contemporaneous year effects, we find similar 18% effect size.
However, effects for all other years are noisy zero.
This is okay as individual year effects can be noisy and its okay for the effect to play out over the entire term limit.
3. There is some evidence of non-linearity in the above relationship.
Using the same specification, but adding a higher order degree 2 asset term in the regression, we get some non-linearity. However, the nonlinearity only kicks in far right part of the support and possibly not that interesting.
4. Number of criminal charges are not related with the bureaucratic churn. Number of criminal charges are also quite orthogonal to the asset levels.
Current/Next Steps
More dimensions can be analyzed here ( incumbent/alignment with the state party, etc). But I am currently focussing on moving one step ahead and compiling data about projects that brought big investments with them. What I have in mind is that assuming big investment projects are decided at higher level than local politician, but local politician can still respond. A big investment project might attract a certain kind of politician because s/he expects that it provides them with more opportunities for holdups. This increases their interference with the bureaucracy and kicks in the phenomenon of interest to me. I am looking into a few potential candidates like list of big projects, construction price index, etc that are likely to be correlated with local economic expectation.
Procuring posting data of more states