Determinants of the duration and ending of terrorist and other non-state armed groups
B.1 Additional Figures
Figure B.1: Map of battle incidents by group
Sources: Author’s calculations. Battle data: Raleigh et al. (2010); Map data for cities: UN-OCHA (2011) and National Geospatial-Intelligence Agency (2013); Map data for Ethiopia and neighbouring features: DIVA-GIS (2013).
127 B.2 Additional Tables
Table B.1: Regression results firm investment, different conflict buffer sizes Dep. Variable Investment Rate, Fixed Effects Regression
Different Buffer sizes 30km 30km 100km 100km
allbattle30km -0.010** -0.008*
Robust standard errors; t-statistics in parentheses. ***,**,* denote significance at the 1%, 5%, 10% level respectively.
Table B.2 : Regression results firm investment, lagged conflict measures Dep. Variable Investment Rate, Fixed Effects Regression Using lagged conflict intensity
No. obs. 2581 2431 2060 2266 1921
No. firms 631 606 540 591 528
R sq. 0.003 0.000 0.022 0.000 0.023
Robust standard errors; t-statistics in parentheses. ***,**,* denote significance at the 1%, 5%, 10% level respectively.
128
Table B.3: Regression results firm investment, different insecurity measures Dep. Variable Investment Rate, Fixed Effects Regression
Effects of other insecurity indicators
No. obs. 2581 2060 2581 2060 2581 2060
No. firms 631 540 631 540 631 540
R sq. 0.000 0.022 0.000 0.021 0.000 0.021
Robust standard errors; t-statistics in parentheses. ***,**,* denote significance at the 1%, 5%, 10% level respectively.
Table B.4: Regression results for other firm indicators
Fixed Effects Regression, effect of conflict on other firm indicators Dep. Var.: No. new workers New workers/
total workers Sales Firm size
(log) Profit rate
Battles 50km -0.208*** 0.011 -0.019 -0.006 0.002
(-3.08) (1.19) (-0.08) (-0.98) (0.21)
Profit rate 0.575* 0.049*** 0.008
(1.72) (2.83) (0.43)
Output variation 0.532 -0.025 4.399 -0.029 0.230*
(0.48) (-0.41) (1.04) (-0.46) (1.83)
Road connection 3.414** 0.112 7.373 0.126 -0.088
(2.08) (1.55) (1.55) (1.27) (-0.68)
Total sales 0.000*** -0.000 0.000*** 0.000
(4.24) (-0.26) (2.61) (1.53)
log Firmsize (Workers) 1.214* 4.600*** 0.035
(1.88) (3.48) (0.43)
No. obs. 1493 1493 2300 2060 2060
No. firms 490 490 566 540 540
R sq. 0.039 0.016 0.011 0.014 0.007
Robust standard errors; t-statistics in parentheses. ***,**,* denote significance at the 1%, 5%, 10%
level respectively.
129
Table B.5: Regression results firm investment in Addis Ababa
Dep. Variable Investment Rate, Fixed Effects Regression Sample from Addis Ababa
Battles 50km 0.004 0.002 0.006
(0.85) (0.36) (0.15)
Profit rate 0.052*** 0.052***
(4.09) (4.09)
Output variation 0.005 0.005
(0.11) (0.11)
Road connection -0.048 -0.272
(-0.54) (-1.13)
Total sales -0.000 -0.000
(-0.40) (-0.39)
log Firmsize (Workers) 0.068*** 0.069***
(3.04) (3.06)
Year dummies no no yes
No. obs. 4271 3297 3297
No. firms 928 776 776
R sq. 0.000 0.018 0.021
Robust standard errors; t-statistics in parentheses. ***,**,* denote significance at the 1%, 5%, 10% level respectively.
130 B.3 Technical Notes
As mentioned in the text, the data on the location of towns was obtained using two different datasets. The first is a data compilation done by UN-OCHA (2011, available at:
14,2013), which is a part of the so-called Common Operational Datasets and gives coordinates for populated places in Ethiopia. The original data was gathered by the CSA, the International Red Cross and the Food and Agriculture Organization of the United Nations.
The data is contained in a GIS Dataset in the ESRI Shapefile format. This data was complemented and counterchecked by a dataset of official (US-American) foreign names published by the GEOnet Names Server and developed by the National Geospatial-Intelligence Agency (available at:
accessed March 14,2013). The data is a text-file containing the name, type and some more information about each listed location as well as the GPS coordinates.
In the firm data a variable indentifies the town only by a number and a list was obtained from the CSA, giving the names of the 107 towns belonging to the numbers. The list contains no further information like regions or other political divisions. Ethiopian town names are problematic because quite often there is more than one town with the same name and there are different ways of transcribing Ethiopian names. To identify the correct town, further information from the firm survey was used, which identifies more detailed political divisions where the firm is located, although that information is not always completely consistent. This additional information was then compared to the different entries in the two above databases, finally identifying the GPS coordinates of all towns except for three (with a negligible number of firm-observations or no observations at all).
The resulting table was converted to a point-shapefile (ESRI shapefile format) using ArcGIS and with the same program a polygon-shapefile containing the buffer-zones of various sizes were produced. The ACLED data is contained in a point-shapefile that identifies the incident locations and contains information about the year and type of incident in the file’s attribute table. The counting exercise was not done in ArcGIS but with a Python script using the Python bindings of the OGR library (part of the GDAL library originally written for the C
language:
library creates an object from a shapefile that provides methods for manipulation and
131
analysis. Among others it allows to set filters according to the geography (spatial extent) or the feature attributes of such objects. This was done with the shapefiles containing the buffer zones and the one containing the ACLED data. Then looping through all years for all event-types, the features falling into each of the buffer zones were counted and the results saved in a SQLite database (this was of course done for all buffer-zones of different sizes).
The resulting dataset could then be merged with the firm data based on the town identifier.
132
Appendix C
C.1 Additional Tables
Table C.1: Regression results organization duration, including political orientation
Nationalist/Sep. 0.579** 0.761 0.740 0.910
Hazard ratios reported; t-statistics in parentheses. ***,**,* denote significance at the 1%, 5%, 10% level respectively.
133 C.2 Technical Notes
The main data source for information about terrorist and other armed groups that is used here, are the Terrorist Organization Profiles (TOP; MIPT, 2008). This database was updated until 2008 when the project ceased to exist. To this point, the organization profiles are made available from the website of the ‘National Consortium for the Study of Terrorism and Responses to Terrorism’ (START) but they are not maintained, checked or updated. The TOP data is not downloadable as a dataset but is hosted on more than 800 separate webpages (one page per organization). The data was gathered from the pages using a self-written web-scraper (written in Python) that extracted the relevant information from each page.
The dataset from Jones and Libicki (2008) is a printed table in the appendix of their study.
Both databases were merged after the names had been manually adjusted (different sources sometimes use different names for the same organization, often due to ambiguous translation possibilities).
Manual additions and changes that are based on the text descriptions in the TOP data, the Global Terrorism Database, the RAND/MIPT Terrorist Incidents Database, the ‘Violent Extremism Knowledge Base’ (VKB) of the Institute for the Study of Violent Groups (ISVG), the research project ‘Mapping Militant Organizations’ from Stanford University, the ‘South Asia Terrorism Portal’ and internet news searches, were made in Stata. A documentation of the about 1400 manual changes and additions in the data is available upon request.
The resulting dataset contained one row (observation) per organization. To merge in the time-varying components it was extended to contain one row for each year an organization existed, resulting in a data-structure suitable for discrete-time duration analysis with one year time intervals.
134 Eidestattliche Versicherung
Ich versichere an Eides Statt, dass ich die eingereichte Dissertation „Micro-Level Impacts of Conflict and the Duration of Armed Groups“ selbstständig verfasst habe. Anderer als der von mir angegebenen Hilfsmittel und Schriften habe ich mich nicht bedient. Alle wörtlich oder sinngemäß den Schriften anderer Autorinnen und/oder Autoren entnommenen Stellen habe ich kenntlich gemacht.
Göttingen, 18.06.2013
Dominik Noe