User:Dcs002/Stampede stats

Thanks!
Thanks to Pldx1 for compiling a huge amount of data to really get this issue stated in reliable, scientific terms.

In a nutshell
Race is the primary predictor of our use of the word stampede in article names and section headings. Common dictionary definitions of the word are applied randomly, though terminology was generally consistent with source material.

Reason for this page
This page was created to show a statistical analysis of Wikipedia's use of the word stampede in article names and section headings throughout Wikipedia describing various types of crowd disasters. All articles and sections were linked from Wikipedia's List of human stampedes (now renamed List of human crushes), so they were already associated with the label stampede at the time of this analysis. (Note: This fact can be a source of Selection bias, as there are some articles about crowd crush and collapse disasters that were not included in this analysis. It's unknown what effect this bias had on the outcome.) All information in this section is based on Wikipedia pages as they were on September 28, 2015.

The goal of this analysis was to assess our use, within Wikipedia, of the term stampede to describe crowd disasters. Earlier work showed that there was a major difference along racial and geographic lines. This is not an attempt at WP:OR. This is an evaluation of the outcome of our current practice of applying the word stampede as each editor thinks it best fits, primarily by use in RS, or by their understanding of the meaning of the word, or for any other reason. This is how Wikipedia is meant to work.

However, several academic sources have raised concerns about the use of stampede in the press, it being a loaded term, implying animalistic and selfish behavior, and shifting responsibility for a crowd disaster onto the victims instead of the organizers of mass gathering events for their failures in crowd management and safety practices. If we determine this word to be laden with such meanings and connotations, and if we are applying it disproportionately to non-white races, then we need to reconsider how we use this word. We are responsible for the excellence of our encyclopedia, not our sources.

This section is meant to summarize in an easily readable manner the findings of a thorough statistical analysis of this question.

Race
The strongest predictor of the use of the word stampede in a title or section is race (which I did my best to define myself as white or non-white). No predominantly white events were named stampedes, while predominantly non-white events were usually named stampedes (73%).

Location
As a check against my subjective classification of events by race, comparisons were again based on geographic group. Group A included the US and Europe, and group B included everywhere else, i.e., Latin America (South and Central America and Mexico), Africa, Asia, The Caribbean, and the Middle East. (These were the groups used in Pldx1's excellent table, which I used as a framework for this analysis. Note that three primarily non-white events occurred in the US.) Similar results occurred. Events in the US and Europe were rarely called stampedes (5%), while events that happened elsewhere were usually called stampedes (74%).

Sudden triggering event
Many definitions exist for the term stampede as applied to humans. Very broadly, these definitions generally require a sudden event or trigger. The presence or absence of a sudden trigger (according to my best determination, including events like fire, release of tear gas, rumor of a bomb, etc.) did not predict whether an event was called a stampede. Of events with a sudden triggering event, 36% were called stampedes. Among events without a sudden triggering event, 53% were called stampedes. These two numbers were too close to be considered statistically different. They are "within the margin of error", as pollsters would say.

Other predictors
Year, religion, and crowd size also predicted the use of the word stampede. Recent events, events of a religious nature, and events with the largest crowd sizes were all much more likely to be named stampedes. However, in recent times there have been more reported non-white events, religious events are more likely to be predominantly non-while, and the largest crowd sizes tend to be non-white. In other words, the contribution of these factors can in part be explained by their connection to race. In a statistical model combining these factors, race comes out as the most important factor.

Conclusion
The use of the word stampede in titles is not related to the meaning of the word, as defined in numerous dictionaries. Use of the word is scattered randomly across articles about crowd disasters, whether or not they have a sudden triggering event. Our use of stampede is primarily and overwhelmingly related to race (as I did my best to define) or whether the event occurred in the US/Europe or other regions. The use of the word "stampede" did, however, generally follow source material, though different sources used for the same event were not themselves consistent in their usage. Even different source articles from the same publications were sometimes inconsistent, with one newspaper publishing an opinion against using the word "stampede" and another article using the word in its headline.

Technical information
This section is intended to show the details of how all of these variables were analyzed and what the specific mathematical significance was of the variables. This will allow critical comments and permit different types of analysis if appropriate. It's written loosely like a scientific paper, with the above section serving as a sort of introduction. There are no citations and no literature review because this page concerns only what is happening in Wikipedia. It is an internal assessment that stands on its own.

Dataset
The starting framework for this analysis was the table compiled by Pldx1. The data in this table was current as of October 9, 2015. The data came from Wikipedia's List of human stampedes, in which a total of 92 events were listed from October 11, 1711, to September 24, 2015. Each case was examined, and information was used from the page versions that were current on September 28, 2015.

Dates
The reasons for choosing September 28, 2015, are as follows: I had originally compiled data from article names on that date and posted it to the talk page of the article 2015 Mina stampede. The following day, one editor created four new articles concerning similar events at previous Hajj pilgrimages, using the word stampede in the name of all four articles. Another editor changed the names of four existing articles concerning disasters in the US and Europe to include the word stampede. (This would add another five predominantly non-white events to the list that use stampede in the title, and three predominantly white events.) A third editor also changed the name of one article from stampede to disaster that same day. Inclusion of article revisions made after September 28 would give undue weight to the actions of two editors (as opposed to reflecting names that had stood unchallenged for a long time or didn't exist at all). Also, the hypothesis being tested is whether Wikipedia has a pattern of race-based assignment of the word stampede, not whether there is a disproportionality now, at this moment, when such matters are in flux due to current events. Article names and section headings were relatively stable prior to September 29, with the primary exception being the article on the 2015 Mina disaster/stampede.

The dates used in Pldx1's table were changed to event year only to accommodate software limitations in using dates prior to the 20th century.

Other variables
The number of dead and injured was updated with sources available on the Internet as of September 28 under the assumption that, if either turned out to be a predictive factor, it would be predictive in the RS material used to edit the articles. An important assumption underlying most of this work is that usage in Wikipedia directly reflects usage in sources. Similarly, crowd size was determined when possible from the article or from other sources.

Events were assigned a type, using my best judgement. Event types included religious, music or music festival, theater or hall, sport event, political - protest or other, retail or giveaway event, other public festival or celebration, war, and fire. Events involving war and fire were placed in those categories regardless of the setting or type of event that was happening. E.g., the Iroquois Theater fire was listed as event type fire, not theater or hall.

Also noted in the dataset was whether the event was indoors or on a bridge, or whether it involved crowd movements in opposing directions (counterflow). The dominant age and sex of the crowd members and those killed were also determined, when possible (men, women, adults, children, women and children, and general - all age groups, both sexes, and all genders). Note: There were no specifically GLBT events on the list, though there were numerous events from which GLBT persons would likely be excluded or compelled to be discrete.

Information about specific, sudden triggering events was also collected. To be considered as a specific trigger, an event would have to be sudden (i.e., rapid onset, e.g., fire, someone shouting "fire" or "bomb", rumor of immediate danger (like a bridge collapse), sudden onset of severe weather, police firing tear gas into a crowd, etc.), and not one that was generally expected or had a gradual onset (e.g., a football game or a church service ends, a crowd moves en masse at their assigned time along an assigned route, a crowd gathers at a location for an expected event, crowd moves in an unpredicted way (like onto a closed bridge or against the expected traffic flow), a stadium becomes overcrowded, etc.). In addition, a specific trigger must impel a crowd to move quickly and with urgency. Failures in crowd management systems were not considered triggering events (e.g., collapse of railings or barriers, overcrowding such that one person falling initiates a progressive crowd collapse, crowd waves, turbulence, or quakes, inadequate ingress or egress facilities for normal flow (like the turnstiles at Leppings Lane at Hillsborough), improper crowd routing, gates opening under crowd pressure, etc.).

In addition to specific triggers, events were classified as escape panics or not escape panics. An escape panic is an event during which crowd members attempt to flee something, such as a fire, tear gas, a bomb (real or perceived threat), or impending physical attack. The goal of the crowd members in an escape panic is to get away from something. This is a different motivation than crowds moving toward a target or goal, such as a temple at the end of an annual pilgrimage, or trying to get the best seats at a general admission concert, or trying to get across a crowded foot bridge to get home after a large public festival.

Finally, lapses and failures in crowd management measures were noted but not analyzed, though they will be at some point.

Statistical analysis
The dependent variable, use of the word stampede in article names or section headings, was entered as 0 if stampede was not used and 1 if it was used, giving a continuous numeric output, allowing for a variety of analyses. Individual comparisons between white and non-white, US/Europe and elsewhere, sudden triggering event yes and no, and indoor and outdoor were made using two-tailed independent sample T-tests, equal variance not assumed unless Levene's Test for Equality of Variances is met. Individual correlations of year, religiousness, and crowd size were determined through simple regression. Simple regressions were also carried out for numeric values of white = 0 and non-white = 1 and for location US/Europe = 0 and elsewhere = 1 to permit ranking of R2 for the different factors. To determine contributions of year, crowd size, race, and religiousness, a multiple regression was performed. A one-way AVOVA with Fischer's PLSD post hoc analysis was performed for age/sex group killed and age/sex group in the crowd. For these analyses, men and women were combined into the adult group after determining there was no significant difference between either group and the adult group, because both groups were very small (n = 1 and 2, respectively). All significance was considered p </= 0.05.

Results
An examination of Pldx1's table revealed 50 events in the English language Wikipedia that were described in their own articles or sections within larger articles on September 28, 2015. Each article name or section heading used either the word stampede or another word to name or describe the event.

Race was the single most predictive factor in whether we use the word stampede in article names and section headings. None of predominantly white events (n = 17) and 73% of predominantly non-white events (n = 33) used the word (R = 0.690, p < 0.001). Geographic group was also a very strong predictor, with 5% of article names and section headings concerning events in the US and Europe using stampede (n = 19), and 74% of articles and sections covering events elsewhere (n = 31) using the word (R = 0.670, p < 0.001). Event year (R = 0.530, p < 0.001), crowd size (R = 0.355, p = .031), and religiousness (R = 0.628, p < 0.001) all correlated significantly with use of the word, but in a combined multiple regression (n = 37 with all variables given), race was the only significant predictor (p = 0.006), though event year and religiousness were marginal as predictors (p = 0.056 and 0.068, respectively). (Use of the term "marginal" for significance of 0.05 < p < 0.10 is not accepted universally among researchers, and rarely would a statement like this be accepted in a research journal.)

Outdoor events (n = 40) were more likely to be called stampedes than indoor events (n = 10, p = 0.049). Traffic counterflow (crowd members moving in two or three opposing directions) and event location on a bridge were not significant predictors of the use of the word stampede. Presence of a sudden triggering event was not significant (p = 0.275), nor was the condition of escape panic (p = 0.477). Incidents with triggering events and escape panics were less likely to be called stampedes, though that trend is not indicative of a trend in either direction.

Of the five events where mostly or entirely children were killed, none used stampede in the article name or section heading. In contrast, all five events in which both women and children made up most of the victims were named stampedes (p = 0.001). (Inclusion in the latter group generally means that sources say most of the dead were women and children.) Among events in which adults were the clear majority of victims, 60% of those events were named stampedes, which was significantly higher than children (p = 0.020), but not significantly lower than women and children. Among crowd disasters in which a mix of adults and children were killed, 42% were called stampedes, which was significantly lower than the women and children group (p = 0.015), but not significantly different from the adult or children groups. The makeup of the crowd itself (adults, children, or both) did not have an effect. Analysis of the gender of those killed or those in the crowd was not possible due to limited data.

Why this is important
The disaster that took place in Mina, Saudi Arabia during part of the annual Hajj pilgrimage stimulated discussion of crowd disasters and the terminology with which we describe them. More than 2,200 people are reported dead, and the issue is an emotional one. There are numerous allegations in source materials that the host government is under-reporting the number dead and attempting to deflect responsibility for this disaster onto others. Others allege that rival nations are inflating the number of fatalities and exaggerating the role crowd management actually played in the disaster. As the Hajj is a religious/spiritual pilgrimage, the role of "the will of God" is always a part of the discussion when disasters happen during the Hajj. So the subject is very important and highly charged in many ways. That makes it especially important for us to describe it neutrally and sensitively (i.e., nothing unnecessarily inflammatory). Our use of the word stampede, though no more than a minor concern to most people, has the potential to serve as a nucleation point for controversy. I think our use of stampede needs to be examined and reconsidered, given the serious racial bias in our use across Wikipedia (the subject of this study), as well as the simple importance of using words according to their current definitions. Both have the ability to bring some level of disrepute onto our encyclopedia, and the racial bias has the potential of offending many people, as well as becoming a PR disaster for us.

Academics and crowd management professionals have spoken out against the use of the word stampede to describe such crowd disasters as the incident in Mina because it does not provide an accurate description of what actually happened (some say misleading), because linguistically it defines the event as a product of the crowd's behavior rather than a predictable outcome of ineffective crowd management, and because it implies animalistic behavior and careless, self-centered actions on the part of the victims of such tragedies. (These opinions are sourced in the article.)

Findings
Our own use of stampede in Wikipedia is not related to most dictionary definitions, which require something sudden or impulsive. Most English dictionaries require suddenness or impulse, though sources can be found that do not. The presence of a sudden triggering event, as described above, did not correlate with our use of the word stampede. The presence of a triggering event was not significant as a determining factor. This is not a precise measurement of adherence to definition, but definitions that are similar will likely yield similar results, as was the case when escape panic was considered. In any case, dictionary definition is only one factor to consider when choosing how we use the word.

The question to be answered on this page is whether Wikipedia has a pattern using a potentially value-laden word in a way that describes events accurately and does not put an undue burden on any group inappropriately, specifically people of non-white races. The answer is no. When all events with article names and section headings as of September 28, 2015 were examined, the only occurrences of the word stampede were in articles and sections describing non-white events, and the word was used in 73% of such articles.

More articles about recent non-white events exist on Wikipedia than older non-white events, religious events are more likely to involve non-whites, and larger crowd size are also more likely to involve non-whites, so the question was addressed whether any of these variables could more suitably account for the uneven application of the word stampede across racial groups. When all four variables (race, event year, religion, and crowd size) were put into a statistical model, the only variable to reach significance was race. In addition, individual analysis of these variables showed that race accounted for more variability in the use of the word stampede than any other variable.

The other factors analyzed may or may not be meaningful to Wikipedia, though some have significant values. When women and children are the victims, we invariably called it a stampede. When only children are the victims, it was never a stampede.

We must consider the likelihood that our use of the word stampede originates in our source material. Objections to changing our use based on findings such as these often cite usage in our sources as a reason to continue as we are. However, if we continue as we are, there is less than a one in a thousand chance that this disparity will go away, according to the statistical analysis reported above. In fact, it's almost certainly far less likely than that, only that is the limit of the software used for this analysis.

Conclusion
Given the expert opinion that the word stampede is laden with negative meaning, and given their assertions that the word is sometimes used as part of a strategy to assign blame to the victims, Wikipedia should not continue to defy dictionary definitions of the word and continue to use stampede to describe something that only (or mostly) non-white people do. This does not make Wikipedia a more excellent encyclopedia. By continuing our current usage patterns, we run the risk of being saddled with charges of racism, and right now, our current practice makes such charges impossible to deny regarding this issue. We have to take responsibility for our own encyclopedia and stop relying on our sources to make us better. If our sources use the word in an incorrectly, racially biased way, that does not excuse us from doing the same. We are doing this knowingly, following an RM on the 2015 Mina stampede page and an RfC on the WP:W2W talk page, both of which had significant participation from editors who are interested in such articles. This is a low probability, high cost risk with a very simple and complete mitigation strategy.

Future directions
Failures in crowd control should be analyzed. These are particularly relevant as it is a common opinion expressed by experts in our RS that government entities (e.g., police, political leaders) and event managers who fail to effectively manage crowds and crowd safety often attempt to shift blame to the crowd itself, using the word stampede as part of that strategy. This is important information when evaluating RS for bias. Wikipedia should weigh its role in propagation of propaganda against use of common names for tragic events. Such strategies are well documented in the West and elsewhere.

Table of data used for this analysis


Notes: Some data have been converted to numerical values. Generally, in such cases, 0 means "no", and 1 means 'yes". In the case of race, 0 means "white", and 1 means 'non-white". The first heading, "event", refers to the event number on Pldx1's table. Under "event type": F = fire; M = musical event; P = public festival, celebration, or other public gathering; PO = political rally or protest; R = religious event; RT = retail or giveaway; S = sports event; T = theater or hall; and W = war. (See text for clarifications and adaptations of these event types.) "Event type num" gives a number for each event type, which was necessary to run a statistical analysis. The number represents only a category, not a value. "Mod type num" is a repeat of event type num. Certain cases needed to be re-assigned to different event types,, e.g., war and retail/giveaway, but all such cases were eliminated from this analysis because they did not have an article or section describing them. Under "event population" and "population killed", the following symbols are used: A = adults, C = children, G = general (combined adults and children), and WC = women and children. (Note: If most victims were children, a C was used. If most dead were women and children, or if the sources emphasized the deaths of women and children, WC was used. "A" was used if only adults were involved.) "Src" and "cat" are taken directly from Pldx1's table (see that page for details). "Stam Y/N" indicates whether the "cat" value was "stam". "Geo" is the same data as the "w" column in Pldx1's table. "Non-white" represents whether the crowd was predominantly non-white. "WP title" indicates whether the word stampede was used in the article name or section heading. Many of these values are subjective, and they represent the best interpretations of the author.