I’m Not That Great a Forecaster: Looking back on my past predictions and learning how to improve

2 Jan

In early January of 2013, I wrote two posts that outlined six conflicts to watch for civilian protection advocates in the coming year.  Without any concrete methodology, I picked out Sudan, Afghanistan, Mali, Kenya, Syria (specifically violence in a post-Assad Syria), and Central Asia.  Arguably, I was wrong in five of the six cases if the criteria is that the violence had to get significantly worse from 2012 to 2013 in the manner that I predicted  (it gets even worse when you think about all 2013 conflicts I omitted; Nigeria, Egypt, CAR, and Iraq all experienced episodes of mass killing that have intensified since 2012).  I’ll briefly outline how I did country by country, address what I did wrong, and because it’s that time of year again, propose predictions for 2014.  If you don’t want to read the whole thing, feel free to skip to the bulleted predictions.

Sudan had a turbulent year, but it’s nothing really out of the ordinary for the troubled country.  #SudanRevolts returned in September and October and prompted a fairly superficial cabinet reshuffle, but not much else.  Violence continued to rage in Darfur, South Kordofan, and Blue Nile.  Although violence increased in Darfur and perhaps South Kordofan, it was not a radical departure for 2012.  In my last sentence of my Sudan section, I briefly mentioned South Sudan.  While the violence in Jonglei between the Lou Nuer and Murle ebbed slightly in 2013, any progress made in the country was obliterated by the crisis that began on December 15th and has enveloped the country.  It’s unclear how many people have died, but it probably constitutes a mass killing. Mali has certainly experienced some violence in 2013, but there is no way that it was worse than 2012.  The French-African intervention was fairly successful at expelling the jihadist-Tuareg alliance from Northern Mali.  Fairly successful elections went ahead and the leader of the initial coup, Amadou Haya Sanogo, has been arrested and will be put on trial.

Afghanistan is probably the only case I got right.  Civilian casualties increased in the first half (and possibly the second) of 2013, marking a reversal in 2012’s trend.  For every success there’s a failure, and Kenya is that failure.  The March elections went off smoothly, and so I was really wrong.  I was right to predict that 2013 would be worse than 2012 for Syrians, but it didn’t happen in the way I thought.  At the time, it seemed very likely Assad would fall, initiating a mass killing of Alawites in and around Latakia.  That didn’t happen.  Instead, massive violence continued against civilian populations throughout Syria mostly with the exception of Latakia.  Finally, like Kenya, I really got Central Asia wrong.  There was not significant violence in any Central Asian country (excluding Afghanistan) this calendar year.  Regime change could have potentially caused conflict, but those pesky autocratic regimes just won’t go away.

So how can I improve?  First, it would have helped if I had had a concrete measurement for what constituted conflict.  Second, it would have made sense to have had a constant metric for assessing why I thought conflict would occur in certain places.  Figuring out what my predicted conflict zones had in common and why they were different from other potential conflict zones would have improved my methodology, even if creating a model from a hunch isn’t the best social science (if someone wants to pay me to blog I promise my methodology will be more robust).  My analysis also would have improved if I had laid out specifically what changes I was looking for and how they fit into a larger historical narrative.  For example, while there was both a history of and a potential for political instability in Central Asia, my only data points were the 2005 massacre in Andijan, Uzbekistan and the 2010 violence in Osh, Kyrgyzstan.

So moving into my predictions for 2014, rather than creating a complete methodology of my own, I’m going to borrow two of Jay Ulfelder’s crowd-sourced forecasting projects as points of reference.  The first is the Center for Genocide Prevention’s online opinion pool (password needed).  The opinion pool aggregates the opinions of currently fifty individuals interested in atrocity prevention to create averaged probabilities about the likelihood of a particular event.  The questions are generally phrased like this, “Before X date, will an episode of state-led mass killing occur in Y country.”  The second is a wiki survey also originating from the Center for Genocide Prevention.  The results demonstrate how much more likely any particular country is to experience an episode of state-led mass killing than other countries during 2014.

In order and with percentages, these are the countries that I think are most likely to experience a mass killing (defined as more than 1,000 civilian deaths) in 2014:

  • Syria (95%)
  • South Sudan (85%)
  • Iraq (85%)
  • CAR (75%)
  • Sudan (60%)
  • Afghanistan (50%)
  • North Korea (50%)
  • Mexico (35%)
  • Nigeria (30%)
  • Burma (20%)
  • DRC (20%)
  • Egypt (10%)
  • Mali (5%)
  • Venezuela (5%)

My predictions are largely based on the crowd-sourced forecasts from Jay’s two projects, aren’t identical.  I’ll explain why, country by country, below.

Obviously, the chances that a mass killing will happen in Syria are very, very high (the wiki survey is definitely wrong in placing Syria 6th).  However, the opinion pool’s average probability that there will be a mass killing of Alawite civilians in Latakia province stands at 43%.  I think this is too high, and the real probability stands nearer 15%.  For a mass killing of Alawites to happen, the Assad regime would have to collapse or severely weaken.  Demonstrated by my false prediction of his doom in early 2013, Assad has proven surprisingly durable.  Civil wars tend to last a long time, so especially with the increasing fragmentation of the opposition, it’s doubtful Assad will be defeated anytime soon.

South Sudan, Iraq, and CAR all have ongoing conflict that will almost certainly include a case of mass killing, committed either by government or rebel forces, in 2014.  Iraq and CAR definitely experienced a mass killing episode in 2013, and South Sudan probably did, but the concrete numbers to confirm it don’t exist.  I pegged the chances of a future mass killing as slightly lower in CAR only because of the combination of the peacekeeping force and the higher potential for resolution than in South Sudan.

Jay Ulfelder, in his review of mass killing in 2013, wrote of Sudan, “…where the uncertainty is not whether the regime is engaging in mass killing but in how many parts of the country at once and targeting how many different groups.”  He’s right, and unfortunately civilians in Darfur, South Kordofan, and Blue Nile will likely continue to suffer in 2014.  In the opinion pool, a question asks the probability that Sudan will carry out a mass killing of anti-government activists will occur in 2015, and places the average at 31%.  I think this is far, far too high.  Despite significant anti-government protests, the body count has always remained low.  The Sudanese regime is intelligent in this respect, instead opting for mass arrests of protest leaders.  The scarcity of bloody street confrontations decreases the potential of igniting the paradox of repression.  It is also telling that the bloodiest anti-government protest this year happened in Nyala, South Darfur.  Khartoum is happy to take the fight to enemies in the periphery, but remains wary of the threat from the center.  If the government were to significantly weaken, there’s an increased chance it would unleash lethal violence against anti-government activists.  However, despite increasing organization from the political opposition and calls that the regime was about to fall, the NCP remains in power.

Afghanistan probably experienced a mass killing with the Tailban as the perpetrators in 2013, and there aren’t a lot of positive signs for the future.  In North Korea, it’s basically impossible to tell how many people are dying each year in giant concentration camps.  50% is simply a wild guess.

Drug violence in Mexico is out of control, but I’m hesitant to peg a high probability of a mass killing because it’s unclear what share of civilians vs. cartel members are killed in drug violence (in a tweet, Jay confirmed that cartel members count as combatants).  Bottom line: a lot of people will die in Mexico in 2014, but it may or may not constitute a mass killing.

Nigeria experienced a classic case of a counterinsurgent mass atrocity last year, and despite some international attention on the issue, there is still a decent chance it happens again.  Nigeria is 14th on the wiki survey for the chances of state-led killing (too low in my opinion) for 2014 but doesn’t appear as a question in the opinion pool.

Burma is a country that is very much in the news for people interested in atrocity prevention, but I’m more optimistic than other analysts about the prospects for 2014.  Burma’s counterinsurgencies against ethnic-minority armies are long-running, but have recently had quite low body counts.  I see no reason for that trend to stop.  The opinion pool predicts there is a 35% chance of a mass killing of Rohingya in 2014.  No single rioting incident has yet escalated to widespread killing (the Meiktila incident killed dozens, not hundreds).  Despite the massive persecution the Rohingya face, the levels of lethal violence have remained relatively low.  Without an obvious trigger, I think more slow-motion ethnic cleaning is far more likely than a full-blown mass killing in 2014.

DRC placed fourth in the wiki survey.  Perhaps this is a bit excessive, but not completely unwarranted.  The combination of a country in transition from autocracy to democracy, the prevalence of armed groups in the Kivus, and intrusive neighbors means the risk of a mass killing in the DRC remains relatively high.

Egypt also suffers from an unfortunate history.  Last year, the government undertook a mass killing in response to pro-Morsi demonstrations.  A similar scenario could repeat itself, violence in the Sinai could rapidly increase, or even less likely, a civil war that involves mass killing could erupt.  All of this is plausible, if not incredibly likely (Egypt is 15th in the wiki survey).

The situation is not absolutely analogous to the DRC’s, but Mali also suffers from a somewhat unstable post-major conflict environment.  The opinion pool average for a state-led mass killing rests at 13%.  I think this is too high (probably suffering from the bias that comes from forecasting rare events) because of the progress Mali has made since 2012, but not terribly so.  In the wiki survey, Mali is 3rd, which strikes me as overly pessimistic.

Finally, Venezuela is a bit of a stretch, but I decided to include it anyways.  Few atrocity prevention advocates are paying attention despite the high political instability and absolutist rhetoric coming out of the Maduro administration.  Though Venezuela appears 82nd on the wiki survey and isn’t in a region of the world that has been prone to mass killing recently, I think a political crisis resulting in a government mass killing is plausible if still very unlikely.

Correction: Jay Ulfelder wrote this in a comment, “One point of clarification about Syria and some of the other cases you discuss: in both the statistical modeling and the wiki survey, we’re looking at the risk that a *new episode* of mass killing will start, not the risk that the one(s) we’re seeing now will continue. So Syria could hypothetically get a very low predicted probability or rank if the models or crowd deemed it unlikely that the state would begin deliberately killing large numbers of civilians from a discrete group it isn’t already targeting now. Hence the question in the Syrian case about Alawites but not one about the groups the regime is killing in large numbers now.”  I didn’t realize that when I wrote the post.

Advertisements

3 Responses to “I’m Not That Great a Forecaster: Looking back on my past predictions and learning how to improve”

  1. dartthrowingchimp January 3, 2014 at 12:21 am #

    Good stuff, Danny. What you’re trying to do in terms of being more specific about what you’re forecasting, how you arrive at your estimates, and scoring yourself after the outcomes are known is exactly what you should do to get better at it. There’s now a lot of evidence that forecasting is a skill we can develop, and this is how you do that.

    One point of clarification about Syria and some of the other cases you discuss: in both the statistical modeling and the wiki survey, we’re looking at the risk that a *new episode* of mass killing will start, not the risk that the one(s) we’re seeing now will continue. So Syria could hypothetically get a very low predicted probability or rank if the models or crowd deemed it unlikely that the state would begin deliberately killing large numbers of civilians from a discrete group it isn’t already targeting now. Hence the question in the Syrian case about Alawites but not one about the groups the regime is killing in large numbers now.

    Also, if anyone reading this is interested in participating in the opinion pool Danny mentions, please email me at ulfelder gmail. And not that it matters greatly, but we currently have almost 70 registered forecasters in that pool, not 50.

    • dhirsch1 January 3, 2014 at 12:49 am #

      Thanks! And that makes a lot of sense on Syria. I’ll post a correction.

Trackbacks/Pingbacks

  1. Looking Back on My 2014 Mass Atrocity Forecasts | The Widening Lens - January 2, 2015

    […] year ago, I took a stab at predicting which countries would experience mass atrocities in 2014 (defined as 1,000 noncombatant intentional deaths caused by discrete group). My predictions were […]

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: