3.5.5 Graphical Representation of Counts

Given counts of violations, an organization may choose to present information graphically rather than in a table. Many people find that graphs are more readily comprehensible than the same data presented in a table.

Consider, for example, Figure 3.5.5.1 below.

Figure 3.5.5.1: Number of cases of detention and torture for one department (province) in Haiti, by month, 09/91-10/94

Note: Violations may not be summed between series because a) they are different kinds of violations, and b) any person may have suffered one or both violations.

Source: Haitian National Commission for Truth and Justice (CNVJ); used by permission.

This figure contains the numbers of cases of detention and torture in one department of Haiti, by month, in the period September, 1991 - October, 1994. The first note at the bottom of the graph observes that a given person may have suffered either detention or torture, or both violations, in a given incident. The information comes from interviews done by the Haitian Truth Commission (CNVJ) in 1995 [13].

From the huge peaks in October, 1993 and September, 1994, it can be quickly seen that these two months dominate the incidence of detention and torture in this department. Political analysts might then reflect on the historical events which might have been associated with these outbursts of violence. A second observation that emerges from this graph is that the two kinds of violence seem to occur at the same times. For example, in the period 09/93 - 01/94, both kinds of violation increase dramatically but then decline to all-time lows in 02/94. Increases and decreases in one series are mirrored in the other series: a tendency to rise in the detention series generally coincides with a tendency to rise in the torture series.

We calculate the correlation coefficient ( r ) between the two series to be r=0.84, a very strong relationship. We have n>30 (n=38) points in the series. For that sample size, the distribution of r is Normal, and the standard error is

in this case, SE( r ) = 0.090. To determine the probability of getting a value as high of r as 0.84 by chance, one calculates the t-ratio, which is simply the value of r divided by its standard error, i.e., 9.027. The t-distribution for that value of the t-ratio gives the probability of getting a correlation of 0.84 or higher if there is no correlation; this probability is usually called the p-value. In this case, the p-value is 0.000000000075. Thus the two series are strongly and non-randomly linked. The connection between the two series leads to questions about why they might be linked. Analyzing the coincidence of two kinds of violence in time I call an analysis of the coherence of different kinds of violence. Note that this is the same technique as was applied in the example presented in 3.2.1.2.

Why are detention and torture so closely connected? Perhaps people who were detained were very frequently tortured. Thus the next question might be in what proportion of cases did both violations occur together? What proportion of all detained people were tortured, and how many cases of torture happened outside of detention? Did these proportions change in time? Did torture happen relatively independently of detention in some periods, while in other periods detention almost certainly meant torture as well? Given data organized as was suggested in Section 3.4, these questions can be addressed. If torture and detention did not occur together at as high a rate as suggested by the correlation, perhaps torture and detention occur at the same months (but not necessarily to the same people) for some other reason, e.g., they might have been two parts of a common policy of repression. At this point we would need further evidence, but given results of highly coherent violence, it is a strong hypothesis.

There are many other ways to count data. The difficult part of using statistics is not generating tables from the data: this is a mechanical application of a database or statistics program's calculation features after repeated data and bad data have been filtered out of the system (i.e., cleaning the data) and the relevant data have been assembled in a matrix. The hardest part of doing quantitative analysis is defining the proper analytical questions, and designing appropriate tables that will best support an argument about a given question. There are other resources that can help human rights organizations with this kind of work [14].