BRAINS is a tool that enables you to isolate the effect of
a specific variable, or a group of variables, on the probability
of having a specific type of boating accident (based on the relationships
between variables in the Boating Accident Report Database (BARD),
which is the universe of all reported accidents).

The Coast Guard does not know if variable relationships of
*non-reported* accidents are similar to those reported in BARD.
BRAINS also gives you the flexibility to make predictions using
variable combinations that are not frequently occurring in BARD.

**What BRAINS Does - by example**

The BRAINS system enables you to isolate the effect that
an individual variable has on the probability of a specific type
of accident occurring (within the universe of all reported accidents).
The benefit of this system is it enables you to measure the
likelihood of a reported accident either increasing or decreasing
given a change in a specified variable while holding the effect
of all other variables constant. Thus, BRAINS allows you to determine
what variables have the greatest impact in determining the likelihood
of common accident occurrences. Further, BRAINS is flexible enough
to allow you to determine the likelihood of uncommon or hypothetical
accident scenarios.

Let's use a specific example to illustrate these points. Assume
that you want to answer the question

**What major factors increase or decrease one's risk of being
in an accident that involves a capsizing, swamping/flooding,
or sinking in New England?**

(For the rest of this example, the model will be simply referred
to as "CAPSIZING").

Further, let's assume that you do not have a copy of BRAINS,
but instead you own a statistical software package and have a
knowledgeable statistician to operate it. This person begins
analyzing the Boating Accident Report Database (BARD) for you.

To answer this question, you first ask your statistician to
compare the incidence of CAPSIZING with two particular types
of boats, open motorboat and canoe/kayak. First, your statistician
reminds you that the results from any analysis of the BARD cannot
report what happened to all boaters - it can only report what
happened to boaters involved in reported accidents.

Now your statistician compares CAPSIZING with boat types.
You discover that in New England:

*Comparison #1:*

- 18% of the open motorboats in reported accidents were involved
in CAPSIZING accidents
- 83% of the canoes/kayaks in reported accidents were involved
in CAPSIZING accidents

Our statistician tells us that **canoe/kayaks are 3.6 times
more likely to be involved in a capsizing accident than an open
motorboat**, i.e., (83-18)/18 = 3.6.

Thinking that water conditions or operator behavior may affect
our results, we inquire further into the accident data. Our statistician
provides us with the following information:

4% of open motorboats involved in accidents reported strong
currents at the time of the accident, whereas 16% of canoes/kayaks
involved in accidents reported strong currents. 3% of the open
motorboats involved in all accidents reported over or improper
loading as a contributing factor, compared to 12% of all canoes/kayaks
involved in accidents.

**Thus, we find that compared with the open motorboats, canoes/kayaks:**

*Comparison #2*

- Are 3 times more likely to be involved in accidents with
strong currents
- Are 3 times more likely to be over or improperly loaded

But this raises a question in your mind regarding capsizings:

**Are canoes and kayaks really more likely to be involved
in capsizings, or are they just more likely to be operating in
strong currents and/or improperly loaded?**

That is, what is more important in affecting the likelihood
of a capsizing: is it the boat type, the water conditions, or
the way the boat was loaded? Using simple statistical techniques
does not seem to provide a clear answer.

When we originally found that canoes/kayaks
were 3.6 times
more likely to be involved in capsizings, we did not consider
that canoes/kayaks are more likely to be in strong currents and
improperly loaded. **This is the major point to consider.**
The original comparison (__Comparison #1__) inadvertently
included the effects of additional variables in comparing the
effect of boat type on the likelihood of a capsizing. In other
words, __Comparison #1__ actually assessed the joint effect
of changes in boat type, water conditions, and load on the likelihood
of a capsizing instead of just the effect of changing the type
of boat. **This automatic inclusion of "hidden" variables
can frequently happen when we use basic statistical comparisons.**

To eliminate these confounding effects, you may have to think
of all the complicated conditions and effects that would have
to be explained to get a clear answer to our original question.
There must be a simpler way...and there is - the **BRAINS**
system.

**BRAINS has done the statistical work for you!**

Using BRAINS, you decide to pick the model __New England
Capsizings__. You will see that the system estimates the
probability of a capsizing given the conditions reported in the
model (the default conditions are the average or most likely
conditions associated with New England capsizings based on accident
report data). You also note that you can change the model by
modifying the value of a variable and, by doing so, can see the
specific effect that a variable has upon the likelihood of a
capsizing while holding the effects of all other variables constant.

You decide to first estimate the probability of capsizing
using the following settings:

**Test Scenerio 1:**

Vessel Length |
16 |

Vessel Type |
Open Motorboat |

Hull Type |
Fiberglass |

Year of Accident |
1992 |

Wind Speed |
7 |

Water Type |
Lake |

Vessel improperly loaded or overloaded |
No |

Stability Loss |
No |

Strong Current |
No |

Probability Result: 13%

These are the BAR variables that are statistically related
to the likelihood of a capsizing. All other variables were found
to be statistically insignificant. Under these conditions, you
see that the model predicts that the probability of a capsizing
is 13%.

**Please Note: The probabilities reported above should NOT
be interpreted as the probability that a capsizing will occur.**

Remember that the BAR data represent reported accidents only.
Although nearly all fatal accidents are reported, only a small
percentage of non-fatal accidents are reported as required by
law. In addition there may be hours of boating activity fulfilling
model conditions that never result in an accident. The models
look at the statistical relationships inherent within the accident
report data. As a result, one should not place emphasis on absolute
predictions; one should instead rely on the relative predictions
provided by the model. These limitations always hold when using
the BAR data and BRAINS.

You now rerun the model changing only the vessel type variable
from open motorboat to canoe/kayak:

**Test Scenerio 2:**

Vessel Length |
16 |

Vessel Type |
**Canoe/Kayak** |

Hull Type |
Fiberglass |

Year of Accident |
1992 |

Wind Speed |
7 |

Water Type |
Lake |

Vessel improperly loaded or overloaded |
No |

Stability Loss |
No |

Strong Current |
No |

Probability Result: 68%

With the result of 68%, what has BRAINS told you? Well, what
BRAINS is reporting is that, holding the other conditions constant
in the model, the specific effect of **a change in vessel type
from open motorboat to canoe/kayak increases the likelihood of
a capsizing by 4.2 times** ([68 - 13])/13 = 4.2)

You now compare this to your earlier estimate (the one done
without BRAINS) on the effect of a change in vessel type which
indicated that moving from an open motorboat to a canoe/kayak
increased the likelihood of a capsizing by **3.6 times**.
BRAINS reported a **4.2 times increase**. Evidently, our earlier
analysis had suffered effects from some additional conditions
we did not account for. BRAINS automatically reported only the
effect of a change in vessel type while simultaneously holding
the effect of all other variables constant.

Let's test BRAINS further. You now want to know what is more
likely to increase one's risk of capsizing; is it boat type,
current, or load? To answer this question, you rerun the model
two more times:

After you set vessel type back to open motorboat, alternatively
switch the other variables (current or load) from "no"
to "yes":

**Test Scenerio 3:**

Vessel Length |
16 |

Vessel Type |
**Open Motorboat** |

Hull Type |
Fiberglass |

Year of Accident |
1992 |

Wind Speed |
7 |

Water Type |
Lake |

Vessel improperly loaded or overloaded |
No |

Stability Loss |
No |

Strong Current |
**Yes** |

Probability Result: 42%

**Test Scenerio 4:**

Vessel Length |
16 |

Vessel Type |
**Open Motorboat** |

Hull Type |
Fiberglass |

Year of Accident |
1992 |

Wind Speed |
7 |

Water Type |
Lake |

Vessel improperly loaded or overloaded |
**Yes** |

Stability Loss |
No |

Strong Current |
**No** |

Probability Result: 86%

BRAINS indicates that among open motorboats, the likelihood
of a capsizing is 42% when there is a strong current and 86%
when the vessel is improperly loaded.

Thus, a strong current increases the likelihood of a capsizing
by 2.2 times and an improper load increases the likelihood of
a capsizing by 5.6 times.

Now we know the answer to our question. Of the three factors
we studied, the following have the greatest impact in increasing
the likelihood of a capsizing:

- Over and/or Improper loading of the boat: 5.6 times more
likely to capsize
- Change in operation from an open motorboat to a canoe/kayak:
4.2 times more likely to capsize
- Presence of a strong current: 2.2 times more likely to capsize