So, what's in National Fire Information Reporting System, this amazing, previously unknown database?
Simply put, everything.
Summarily: time, date, address, time call received, time dispatched, time arrived, time left the scene, number of rigs, which station, which apparatus, type of fire, locations within the building, cause, number of injuries and deaths to civilians and/or fire personnel, even the amount of water and length of hose that was used. House fires, car fires, grass fires, trash fires, false alarms, investigative calls. If there were a line for a firefighter running in place, it would be in there.
CAR dream
It's a CAR reporter's dream. It can also be his or her worst nightmare.
The data field code book is almost two inches thick and varies slightly from state to state. Some fields cancel each other out and others are completely unreliable. What at first glance appears to be a conflagration can quickly turn out to be a flaming paper bag. It has taken me four years to master the database because of its complexities. Best said, there's too much information, and in the hands of a haphazard reporter, it can easily turn into an ombudsman's career. It is extremely easy to make a mistake.
As with most databases, the information is entered by hand. Since 99 percent of the entries are numeric, keypunch accuracy quickly goes out the window. While the generic numbers the number of house fires, the number of false alarms, etc. - are normally reliable, the more specific information should always be double checked on the database and with department records if you can get them.
Specifics
The information is specific. A house fire can be further detailed to show if it's a two-story, brick, no-foundation house. Buildings are detailed by their use, such as a bowling alley, a 22-unit apartment, a nursing home, an elementary school, or a church.
Addresses should always be confirmed through the department or in a city directory.
Firefighters often are reliable when it comes to dousing flames, but filling out reports is another matter. For example, the address of a burning house at the corner of an intersection will either be entered accurately or as "0000ELM" or as "ELMandMAIN," entries that make mapping difficult at best.
Car fires have no address so the entry may either be the intersection, the street block number or the address of the nearest house. Assuming the latter, the information at first appears to be a house fire. The car fire information is found in a different data field, and its type whether it was an overheated engine, a smoldering seat cover or a true car fire - in yet another.
Smoke detectors
Last year, we analyzed the effectiveness of the Detroit Fire Department's two-year-old smoke detector giveaway program. The department didn't have the addresses to which they provided the detectors, but did have a tally by Census tract. Since the database also includes Census tract numbers, the story was still a go.
By isolating the number of actual house fires (you have to combine three data fields to assess the validity of a fire), separating the fires at abandoned houses, and adding the data fields that provide smoke detector information (there are three fields showing whether a detector was present, whether it worked and what role it had in warning occupants about a fire), we were able to tell which areas of the city had a greater number of detectorless fires and, arguably, the greatest need.
The database includes damage assessments, so we further isolated those fires that had significant damage. If there was little damage but a civilian injury, the information was included. Houses where there was more than one fire in a year were counted as one each year.
Other data faults
Then we grouped the information by Census tract, and to our dismay found scores of Census tracts that did not exist. Why? The department had been entering tract data based on indexes from 1980, not 1990 when several tracts were melded or separated into different tracts. We were forced to use a current city directory to fix the tract data and used a computerized tract index provided by the U.S. Census Bureau to clean the information.
Since the department said it based its giveaway on need, we used Census data to profile the tracts that received the detectors and those that had the greatest number of detectorless fires and greatest need. What we found was telling: The department often played politics with a $15 smoke detector by providing them to areas that not only needed them least, but could afford them most. The areas that had the greatest need and often the most injuries received nothing.
To personalize the story, we isolated the detectorless fires that had an injury and crossed them with another database that provided the names of the injured (called a "casualty report," you must ask for this separately from the NTIS or state fire marshal). We tracked the victims, and told their story.
The department has since changed how it gives away smoke detectors, many of which are donated by large corporations. The National Fire Protection Association also has adopted our method as an effective way to assess smoke detector need, and recommends it to participating departments.
In East St. Louis, we found how unreliable address ownership data can be. There are two data fields - owner and occupant - that are generally blank, wrong or interchanged. We needed to isolate the fires at houses owned by the two land baron families. We did this by hand checking every address with the county's tax rolls (the county had provided us with useless data tapes on three separate occasions, and we opted to do it by hand rather than waste any more valuable time). Once done, we were able to assess that the ratio of fires was greater for the land barons than in the rest of the city, indicating something more sinister might have been occurring.
Try it out
As you can see, the database is swamped with information. Some innocuous, some vital. Most is accurate; much is not. The only way to know is to get the database and play with it. Start with a simple story, such as the number of arsons (two codes fit the bill: suspicious and deliberately set) or the number of false alarms. Danger zone: These are initial reports. Get the final cause for every apparent arson from the local fire marshal; otherwise the data will be skewed.
Trying to tackle a much more difficult and complex piece, such as whether slumlords are burning their houses or whether the fire department effectively uses its resources, will do nothing but intimidate and frustrate you and cause you to fear what could be one of the newsroom's most valued resources.
If you run into a problem (which is likely), call the data analyst for the state fire marshal. They are the masters and often will be only too happy to help.
David Migoya can be reached at (313) 222-6677, or send e-mail to migoya@det-freepress.com, or to 75613,262 (CompuServe).