76
edits
Changes
Jump to navigation
Jump to search
→Analysis Stage 1
*The size and significance of the correlation is sensitive to number of deals vs the amount of funding. Results are more significant for number of deals compared to the results for value of the deals.
*Including SBIR data significantly reduces the number of groups in the panel and hence the sample size.
==Analysis Stage2 ==
All the final data files related to Hubs analysis are located here:
Z:\Hubs\2017\hubs_data
*The stata do file '''hubs_load_data.do''' uploads the final hubs data file. This is the raw data that includes variables early funding, late funding, number of early funding deals, number of late funding deals, dummies for whether a hub is present etc. The do file processes data, creates more variables. It also estimates the hazard rate model, and matches the sample based on estimated hazard ratios. It eliminates all city_states that were never matched with a neighbor in ANY year from 2000-2017. The output from this file is '''matched_hubs.dta'''.
*We need a panel data with controls and treatments, such that for each matched control-treatment pair, with pairid = n, in '''matched_hubs.dta''' we assign the same pairid = n to observations of city_states 2 periods before and after they were assigned control and treatment status originally. The matlab code in '''time_matching_hubs.m''' performs this. However the resulting panel has problems of overlap. For instance if chicago_2008 was treatment and boulder_2014 was control with pair id 200 in the original hazard rate match, chicago_2009 is a control in some pair 203 and this observation gets matched with our pair id 200. Need to resolve this.
