Carefully interpreting climate-conflict results (JPR 2012) - Part 2

This is the second comment on articles in the recent climate change special issue of the Journal of Peace Research.  My first comment is here.

Four out of the sixteen articles in this special issue are case studies: Feitelson et al. (Israel-Palestine), Bernauer & Siegfried (Central Asia), Benjaminsen et al. (Mali, discussed here) and Adano et al. (Kenya).

I won't evaluate these studies in detail, but when interpreting their results, we should be mindful of the general likelihood of obtaining false-negatives if we use case studies to answer the question "Does the climate influence conflict?"

Using our own work on civil conflicts as a benchmark, consider that the likelihood of new conflict in a randomly selected country in a randomly selected year is 3%.  That means in 97% of country-year observations, there is no new conflict.  We estimated that 20% of conflicts (globally) were influenced by the El Nino-Southern Oscillation (ENSO).  This means that 0.6% of country-year observations should exhibit ENSO-related conflicts (since 20% of 3% is 0.6%), or alternatively, that 99.4% of country-year observations should not exhibit ENSO-related new conflict, (even though there is a strong and clear association between the two in large samples).

What this means is that if I were to try and detect the relationship between the ENSO and conflict using a case study for a few years in a randomly selected country, I would almost certainly observe "no ENSO-related conflict" (in 99.4% of years).  But this wouldn't necessarily be the right conclusion.  It's just the conclusion we would be forced to adopt if we only could observe one country for a short period.  But if we adopted this conclusion,we would be wrong: this null result is an artifact that emerges because conflicts are generally rare events, and only a fraction of conflicts are related to the climate.  Thus, in order to get a reliable "yes" or "no" answer to the question of whether a rare event (conflict) is affected by external forcing (climate), we need to observe many populations for many sequential periods.

If this unclear, imagine that we want to ask the scientific question: "Do whales exist?" If I went out in a row boat for a day and did a case study of whether there were whales just off my local coast, then I would probably not see any whales.  This would be a perfectly valid observation, but I would be wrong to conclude "whales must not exist" just because I didn't see any at the specific place and time when I was looking for them.  Whale sightings are rare events and we need to watch many locations for many days to have a good sense of how many whales are out in the ocean.

To be clear, I am not criticizing case studies as a method to understanding what is going on in a specific location in a specific point in time.  And I am not criticizing the case studies in this issue.  All I am saying is that if we read a few case studies that report "nothing to see here," and we are looking for rare events, then it is incorrect to jump to the conclusion that whatever rare event we are looking for does not exist.

No comments:

Post a Comment