Obstruction of OSCE/SMM, May 2015 – indicators, charts, dependencies

Coding and construction of a weighted additive index in order to measure the degree of obstruction of OSCE/SMM’s freedom of movement and restricted access by parties in the Ukrainian conflict. Application with data of OSCE reports for the period of May 2015. Context (1), data (2), coding (3), index (4), graphics (5), experimental (6).

If you are familiar with the matter, you should be able to get the idea by looking at the pictures and tables.

1. Measuring obstructions of OSCE observers

In February 2015, leading figures of the Ukrainian conflict agreed on a plan to de-escalte the armed conflict and to create an environment that allows the elaboration of a sustainable political solution. This plan, a bundle of measurements is called ‘Minsk 2′. OSCE/SMM was appointed to monitor the process of its implementation. The agreement includes freedom of movement for multinational observers.

Despite this agreement, conflict parties regularly restrict access to observers. Aside of observations on ceasefire violations, OSCE/SMM report on these obstructions. This post suggests a way to measure or quantify these obstructions based on data from OSCE/SMM. Why?

a) Providing a descriptive and visual overview of how parties obstruct the observers’ mission, as well as how often and at which location this takes place. This helps to recognize general tendencies over time. If there are any.

b) One can make assumptions on the reliability of OSCE/SMM data on ceasefire violations. – The more observers are obstructed in their function, the more distorted is their data on ceasefire violations. In case a party obstructs observers more than the other, or if obstruction is more severe in a certain location than in others, this would distort data on ceasefire violations and could lead to a political or geographical bias. This would not be nice.

c) Analytically and due to quantification of incidents, dependencies can be discovered and tested, such as correlations between quantity and quality of obstructions and ceasefire violations by party, regions or ammunition. Such insights might be helpful for building models for the estimation of conflict dynamics or warfare computer games.

2. Types of obstructions

The most common types during May are listed below, beginning with less serious obstrucitons of observers, ending with more serious ones. Excerpts from OSCE/SMM reports are added for illustration of categories of incidents. (See here for daily reports of OSCE/SMM)

Vehicle halted, pass

“The SMM was stopped at a “”DPR” checkpoint located north of Shyrokyne; the SMM had to wait while the “DPR” members requested permission from their supervisors to allow the SMM to continue. After 25 minutes, the SMM was able to proceed without any escort.” (OSCE/SMM, June 12)

Vehicle halted, ID check / searched

In Krasne (government-controlled, 47km west of Donetsk) the SMM was approached by Ukrainian Armed Forces personnel who requested from SMM IDs, the red OSCE booklet and passports. The SMM was held for 15 minutes.” (OSCE/SMM, May 29)

Vehicle halted, escort

“Close to government-controlled Lebedynske (16km east-north-east of Mariupol), members of the Kryvbas volunteer battalion serving under the Ministry of Internal Affairs stopped an SMM patrol at a checkpoint, saying the patrol could not enter the village without an “escort”. After 15 minutes, the commander of the volunteer battalion arrived and “escorted” the SMM through the village.” (OSCE/SMM, June 11)

Vehicle halted, adviced to return due to ongoing fire

““LPR” armed personnel at a checkpoint near “LPR”-controlled Brianka (49km west of Luhansk) asked the SMM not to proceed to “LPR”-controlled Kalynove (58km west of Luhansk), saying the road was mined by a diversionary group. The SMM returned to Luhansk city.” (OSCE/SMM, June 11)

Vehicles halted, access denied

“In “LPR”-controlled Slovianoserbsk (28km north-west of Luhansk), the local “LPR” “military commander” told the SMM that it was not allowed to proceed to “LPR”-controlled Sokilnyky (38km north-west of Luhansk) without further permission from his superiors. The SMM was not allowed to proceed.” (OSCE/SMM, June 13)

3. Coding the degree of obstructions

The severity of obstructions is coded in a ordinal scale, the higher the score, the more observers are impeded in their function. The criteria for coding is the assumed loss of time and the limitation of freedom of movement. The definition of values per category and relative differences are not derived sufficiently, they are a priori. Observers themselves would be at the best place to make assumptions about the value of parameters – and they are most welcome to do so.

!Degree of obstruction (d)
d = 1 if vehicle halted
d = 2 if vehicle halted, ID check / searched
d = 2 if vehicle halted, escorted
d = 3 if vehicle halted, ID check / searched / mobile phone removed
d = 4 if vehicle halted, adviced to return because of ongoing fire
d = 5 if vehicles halted, access denied

Additional notes on coding of incidents:

  • Data is based on daily reports, published by OSCE/SMM (see link above). At the end of each report all obstructions by conflict parties per day are summarized. Only information in those summaries are coded.
  • Obstructions are coded by time (day of incident) by party (government including independent armed groups, separatists), location (each incident is assigned to the area around Mariupol, Donetsk or Lugansk, depending on geographical distance as mentioned in OSCE/SMM reports).
  • If it was observers’ own decision not to enter an area because of mines or ongoing fire, the incident is not coded as an obstruction. If they were advised by conflict parties to do so, it is coded as an obstruction.

4. Weighted additive index (s)

A weighted additive index (weighted sum model) combines quantity and quality of obstructions. It’s defined as the sum of incidents (o), multiplied by its degrees (d) as defined above. I think in its general form it looks like this:

S_i^score = sum from { { i = 1} } to { n } d_i o_ij)

Fig. 1 is a screen shot of the spread sheet. It illustrates coding of obstructions in DPR controlled areas around Donetsk for the first two days in May as well as the calculation of s. The figure shows some additional categories, they should be self-explanatory. They are not relevant for the period of May. s is noted as ‘score/3’ or weighted ‘incidents’ in the graphics.


Fig. 1 Composition weighted additive index – screen shot

5. Comparing indicators and parties, obstruction of SMM/OSCE observers in May 2015

Selection of charts on obstructions of SMM. (The spread sheet can be downloaded here as .gnumeric (original) and .xls file. Data on obstructions are located in sheet 4. The rest of the file includes data on ceasefire violations. In the beginning of the first sheet is a brief overview for navigation within the data file.)

Fig. 2 compares the sum of incidents (o) with the weighted sum (s) by parties. s is divided by three for a better comparability with the amount of incidents. (The factor is a priori and might require further calibration.) Same settings by regions in Fig. 3. It’s worth noting the relatively high amount of cases in which the location of incidents is not mentioned in the reports. Those incidents took place while observers tried to inspect artillery sites. Fig. 4 compares s and o over time, by parties and regions.




The overall outcome of the two indicators s and o is quite similar. This is also reflected by a high correlation between both indicators (r=.93). The main difference in their effect is that s ‘punishes’ complete restricted access more than o.

6. Experimental

! The following is experimental. It should also demonstrate the possible application of data. It is no valuable contribution to contemporary political discussions.

Looking for patterns between s, o and the amount of ceasefire violations (cv). (Data for cv are taken from a previous data collection sessions and are included in the spread sheet, it was presented on this blog some posts ago.) Testing for cv (t-1, t0 and t+1). Considering a concrete situation, one will find various possible reasons why and how cv is connected with s and o. Or not.

Fig. 5 provides a visual comparison of cv (artillery and non-artillery) and s for May 2015. The idea of s – o in the lowest chart: the higher s – o, the more severe are single obstructions. Conflict parties take into account a higher degree of defection towards OSCE. This might reflect a temporary change of perceptions or priorities of conflict parties’ personnel (towards SMM). This change of perception might go along with changes related to the military situation. Or maybe not. (At the end of the post you find another version of Fig. 5 with periods marked when s – o is positive for more than a day.)


Pictures can be misleading. So let’s have a look at the correlations at Tab. 1. (Due to the nature of data material there are no significances to expect, aside of the correlation between s and o, but let’s assume there is some relevance.) The amount of ceasefire violations (cv) is estimated minus and plus one day. This is a simple way to test for time dependencies. – In case todays obstructions correlate with tomorrow’s ceasefire violations, then the present would explain the future. (Note: there are way more sophisticated methods for looking for time dependencies. This should provide a first glance in order to find interesting dependencies for further research.)

As you can see in Tab. 1 I constructed an additional indicator (o*(s-o)). The idea of this indicator is to increase ‘spikes’ in the data in order to increase its explanatory power for cv (t-1). Consider that all figures presented in this section (6) is a selection, consider as well that correlations of data do not mean casual dependencies of real events.

Tab. 1 ceasefire violations vs observations, May 2015 – (time) dependencies

Correlations cv (t-1) cv (t) cv (t+1) o s o*(s-o)
cv (t-1) 1
cv (t) 0.135 1
cv (t+1) −0.335 0.122 1
o 0.244 −0.004 −0.187 1
s 0.178 −0.080 −0.216 0.936 1
o*(s-o) −0.078 −0.264 −0.063 0.416 0.67 1

So I started playing around with the data, conducted checks on regressions and under consideration of time dependencies. In short, after comparing the various models, it seems that the indicators are sensible in respect of time manipulation. o*(s – o) is a imprecise variable but there might be some explanatory power (it contributes more to the explanation than creating chaos in models).

Tab. 2 shows the ‘model’ that was most interesting to me. (Also considering limited auto correlation by not having s and o in the same model.) While s and o show similar signs on models cv (t-1), the two indicators show different signs for models cv (t, t+1). This difference is interesting, but it could be also the result of low amount of cases and the still ordinal character of data.

Tab. 2 regression, ceasefire violations vs obstructions, May 2015 – best of all bad models

SUMMARY OUTPUT Response Variable cv (t-1)
Regression Statistics
Multiple R 0.321
R^2 0.103
Standard Error 8.780
Adjusted R^2 0.036
Observations 30
df SS MS F Significance of F
Regression 2 238.481 119.240 1.547 0.231
Residual 27 2081.519 77.093
Total 29 2320.000
Coefficients Standard Error t-Statistics p-Value 0.950 Upper 95%
Intercept 10.403 2.684 3.875 0.001 4.895 15.911
s 1.882 1.103 1.707 0.099 −0.381 4.145
(s-o)*o −1.176 0.805 −1.461 0.155 −2.828 0.475



*Comments, critics and insults are most welcome, especially contributions (such as guest posts) from people familiar with statistics, econometrics or computational modeling. Most interesting might be time series models with fancy stuff like included splines and the like.