Sample Math IA
Sample Math IA
Chunyang Ding
Mr. Kessler
14 December 2012
The year is 2016, and you are an American long jumper contesting to compete in the Rio
de Janerio Olympics. You know that regardless of what happens, you want to do your very bets,
but it sure would be really comforting if you were able to predict the height that would net you
the gold medal. When it comes down to the moment, when the entire country is watching you,
and when the world focuses their eyes on your final Fosbury Flop, you need to be one hundred
percent prepared, both physically and mentally. You HAVE to live in that moment; it is the only
One of the ways that any high jumper can better prepare themselves is by knowing the
competition, or at least knowing what the competition should be able to do. Unfortunately, all
previous attempts to spy on the Russian and Chinese team’s practices have led to a speedy
eviction, usually with several large dogs following behind. Another way that you could
potentially outmatch or outwit the competition is by studying the trends of Olympic high jumps,
and from that, figure out the minimum height that you should be reaching.
To do so, you find data of previous Olympic high jump records from the International
Olympic Committee, and begin processing the data. In order to predict the gold-medal height for
this year, one should try to find a general trend of the data, and use that model. From the model
should emerge the general trend of the high jumps through the year, revealing a better
Initially, the only data you find is from your coach’s old notebooks, dating from 1932 to
1980. This data is shown below, with a slight modification. In order to simplify the regression
process for the trend, it is better to not take the entire year in account, but only the number of
years passed since 1900. This provides a good baseline for your model and would be easy to
understand.
in cm
32 197
36 203
48 198
52 204
56 212
60 216
64 218
68 224
72 223
76 225
80 236
The first thing that we notice is that as the years progressed, the gold-medal height also
increased. This shows a clear positive correlation between the years passed and the height
achieved. Thinking about the event, you realize that this does in fact make sense. Every year,
Ding 3
competitors come out of the event with new ideas of how to train and better ways to safely
improve their body. Everybody wants to beat the last year’s record, and would train as much as
possible to do so.
In order to better visualize what kind of trend existed between the data, we plot the data
with the years on the x-axis and the maximum height on the y-axis, as we believe that the
240
235
230
Gold Medal Height (cm)
225
220
215
210
205
200
195
0 10 20 30 40 50 60 70 80 90
Years (since 1900)
In order to determine what kind of line would best fit the data provided, all sorts of
functions should be looked at. There are many families of functions, but because the positive
correlation has already been identified, we can eliminate several families, such as any inverse
functions or sinusoidal functions. The most likely functions remain as either linear, quadratic, or
Ding 4
power functions. Each case should be studied and compared in order to reason which would most
fit the data, as well as which one would make the most sense.
( )
where represents the slope of the line and is the y-intercept. By just looking at the slope
between the first and last points, we can easily find a rough estimate for the slope of the entire
and our points are ( ) ( ) for the minimum and maximum heights, respectively,
we can reason that the slope of the best fit line should be
While this is by no means a perfect slope, it does provide us with a good estimate for the
Using the information, our equation has only one more variable left: the y-intercept.
Finding this variable is extremely easy, as the only step required is to substitute any data point
into the partial equation and solve for the variable. Using the point ( ), we can discover:
Ding 5
( )
( )
Now, we have a full equation for the approximate line of best fit, as shown below
( )
260
Linear Fit
240
220
200
Height (cm)
180
160
140
120
100
0 10 20 30 40 50 60 70 80 90
Years since 1900
Our graph is not extremely accurate, as there are many points both above and below the
line of best fit. However, we can see that this line is relatively close and models the correct
Ding 6
correlation of data. Using our technology, we are able to generate the actual line of best fit, and
240
Linear Fit + Linear Regression
230
220
210
200
Height (cm)
170
160
150
140
30 35 40 45 50 55 60 65 70 75 80
Years since 1900
As you can see from the graph, there is extremely little difference between the two trend
lines, even though the equations of the actual regression, ( ) , are off by a
bit. This goes to show that our original analytical regression is not all that much off of the correct
regression.
However, we could tell from the graph that the line is likely not the best fit possible.
There seems to be a considerable area around the start of the graph that is below our best fit line,
as well as several points that seem to hover above the line of best fit. Later on, we will evaluate
Ding 7
exactly how well does our data fit, but for now, let us explore a different type of function that
It seems that it is possible that our graph actually displays a trend in the form of ( )
regress that by hand properly. Currently, at this level, we do not have the necessary tools in order
to work out this regression. However, programs such as Excel or Mathematica do have that
capacity, and when the data is imputed, the following power graph emerges.
( )
240
Power Regression
190
y = 103.87x0.1791
R² = 0.8225
140
Height (cm)
90
40
-10
0 10 20 30 40 50 60 70 80
Years since 1900
One additional model that we could compare is a quadratic model. Even though this
seems to be somewhat similar to the power regression, it may in fact be able to better model the
data, given that there is a slight upward curve in the current data. This matches with the parabolic
Ding 8
shape of a quadratic graph, so we can try to apply this regression. Excel reveals the following
( )
240
Figure 1.7
230
220
210
200
Height (cm)
190
y = 0.0113x2 - 0.5075x + 202.67
R² = 0.924
180
170
160
150
140
30 35 40 45 50 55 60 65 70 75 80
Years since 1900
Finally, another model that we could potentially test is a logarithmic model, which tends
to taper off as the x values grow larger. This model could potentially make sense in the real
world, because it doesn’t make very much logical sense for human jumping patterns to
continually grow; eventually we will reach a limit to what our bodies are capable of. Through
( ) ( )
240
Logarithmic Regression
230
220
210
200
Height (cm)
y = 38.194ln(x) + 60.158
190 R² = 0.8156
180
170
160
150
140
30 35 40 45 50 55 60 65 70 75 80
Years since 1900
Currently, we can overlay all four of the previous graphs in order to interpret which regression
may best match the function. However, when we do so in figure 1.7, we may notice that all of
the functions seem to match the region quite well. They generally have about the same number
of data points above the curve as below the curve. Therefore, in statistics, the best way to
calculate how well our function matches our data is to find the R squared value. This process
∑( ( ))
Ding 10
∑( ̅)
̅ ∑
Even though this process is quite convoluted, it is possible to process our data from these
standards. However, for the sake of accuracy, it will be better to process the R squared term with
Excel. This reveals that the quadratic regression is the best way to fit the data, with an R squared
value of 0.924.
However, the information provided by the quadratic equation does not truly make sense.
It would imply that as time progresses, the height that people can actually jump would increase,
and that increase in height would also grow. Eventually, people will be able to jump over 3 story
buildings, or otherwise jump to ridiculous heights. Therefore, for practical purposes, the
quadratic fit and the linear fit, to some extent, does not always make sense. However, for the
small time period that we are observing, it is possible to use these regressions to predict the
heights for years in the same neighborhood as the ones we currently look at.
Using the most accurate model, the quadratic model, it is possible to predict with some
accuracy what the gold medal heights for 1940 and 1944 were. Our x-variable would be 40 and
( ) ( ) ( )
( )
Ding 11
( ) ( ) ( )
( )
These are reasonable results, as they are increasing, as much of the rest of the data would
indicate for them to be, and they do not seem to be increasing extremely rapidly, skewing the rest
of the graph.
Later on, we find more information about other gold medal trials in the Olympics. These
new data points allow us to refine our original model, as if we have more data, our model should
be able to match it. Shown below is previous quadratic model overlaid on the new data, as well
as the new trend line that Excel has processed for us.
300
Additional Data Quad Regression
250
200
Height (cm)
50
0
-20 0 20 40 60 80 100 120 140
Years Since 1900
As you can clearly tell, there is an extremely large discrepancy with what our previous
best fit line is as compared to the current best fit line. For starters, our best fit line has a parabola
Ding 12
opening up, while this new parabola is opening downwards. It may be especially confusing as to
why we had such a large difference, but when we take into account the minimal amount of data
we began with, as well as the difficulty inherent with modeling human behaviors, the best fit
The new line of best fit also makes more sense to us, as it reveals the slowing of growth
as years go on. Although it seems to reveal that eventually, there will be a maximum point after
which humans aren’t able to jump any higher, and in fact start to lose jumping ability, it may in
fact point towards the gradual leveling off of the heights, which makes sense in a physical way.
Using this new graph, we again predict the heights for the “missing years” of 1940 and
1944, as well as the year 2016, as that was the original intent of this portfolio. Doing, so, we get
( ) ( ) ( )
( )
( ) ( ) ( )
( )
( ) ( ) ( )
( )
Ding 13
As we can tell, there are large discrepancies between our original estimates and our new
estimates, but as they are still within a remarkably close range, it is safe to conclude that our new
Through looking at the data, the idea of modeling real-world data has been explored.
However, it must be kept in mind that these functions do not make perfect sense. In the real
world, especially in events with as much error as a high jump, it is extremely difficult to find a
mathematical justification for patterns. For example, if a new type of jumping was invented, or if
better shoes were created, the data would immediately be skewed based on those variables. Also,
depending on a certain person’s body composition or genetic discrepancies within people, there
may be sudden increases due to genetic benefits, or other skews within the data. Additionally, the
Olympic results are more of a result of how much people train than to how the data has shown.
A possible example within our data is the odd discrepancy that occurs at 1948. This may have
occurred because the athletes have not competed for a large number of years, or that new athletes
did not have the needed experience at the Olympics to do well. Whatever the reason, it creates a
problem for the regression, and can largely skew the data.
In reality, there are many variables limiting the height people could possibly jump to.
These variables would not make sense to just increase as time progresses, as that would again
imply that eventually people will be able to jump and fly. There should be an eventual leveling
off, as seen in our logarithmic graphs. Therefore, any graph would eventually fail in its
predictive power. However, for a region surrounding the region being regressed, we can be
reasonably certain that this model would hold, thereby giving the user a reasonable guess for the
With all of this mathematical work done, you have found the expected height of the gold
medal height of the 2016 Summer Olympics to be 247.25 cm. Determined, you set yourself on a
strict training regiment, and when the time comes, succeed in doing so well that you actually go
over the expected value, and hit 253.16 cm. Although you might shrug and think about how no
model is 100% accurate, as it can be so easily influenced by a variety of human factors, this is
not the time for that. Instead, it is the time to revel in your success of taking home the gold medal!