ACC-104: MANAGEMENT SCIENCE (3) Moving averages are most useful for short-term 18 + 28 + 30

N 16 = 𝟐𝟓. 𝟑
forecasting, and may not be as effective for longer- 3
LESSON 5: term forecasts or in cases where data has irregular D 14
16 + 18 + 28
= 𝟐𝟎. 𝟕
patterns or is subject to sudden changes. 3
2ND SEMESTER | A.Y. 202Y – 202Y
LECTURER: MS./SIR
— In the month of July, 45 is too far from 200.
3-mth Moving Average — The limitation in moving average is that the past
FORECASTING Four Forecasting Methods values are still incorporated. Even if there are
Month Actual 3-mth MA changes, it is still not caught up.
 Forecasting is important because we want to Time Series: J 10
forecast demand. F 12
 We want to be prepared as we don’t want our goods 1) Moving Averages M 13 5-mth Moving Average
to spoil. 2) Exponential Smoothing 13 + 12 + 10
3) Trend Projections A 16 = 𝟏𝟏. 𝟕 Month Actual 5-mth MA
 We don’t want to be out of stock or to overstock. 3
16 + 13 + 12 J 10
M 19 = 𝟏𝟑. 𝟕
Example: 3 F 12
Causal Forecasting: 19 + 16 + 13 M 13
J 23 = 𝟏𝟔
Normal Year 3 13 + 12 + 10
1) Regression Analysis (y = {X1, X2, X3…} 23 + 19 + 16 A 16 = 𝟏𝟏. 𝟕
 An important tool J 26 = 𝟏𝟗. 𝟑 3
3 16 + 13 + 12
26 + 23 + 19 M 19 = 𝟏𝟑. 𝟕
 e.g., CQPA / 1 – 5 = f (math score, English score, A 30 = 𝟐𝟐. 𝟕 3
3 19 + 16 + 13
and High school average)—IQ is not a significant 30 + 26 + 23 J 100 = 𝟏𝟔
variable. S 28 = 𝟐𝟔. 𝟑 3
3 100 + 19 + 16
28 + 30 + 26 J 200 = 𝟒𝟓
 Demand for electricity GwH) = function of several O 18 = 𝟐𝟖 3
3 200 + 100 + 19 + 16 + 13
variables (temperature, occasions, population, 18 + 28 + 30 A 350
N 16 = 𝟐𝟓. 𝟑 5
GDP) 3 = 𝟔𝟗. 𝟔
16 + 18 + 28 350 + 200 + 100
 94% of change of demand of electricity = Gross D 14 = 𝟐𝟎. 𝟕 S 28 = 𝟐𝟏𝟔. 𝟕
3 3
Domestic Product (total value of goods and 28 + 30 + 26
O 18 = 𝟐𝟖
services created within the country) — We need 3 months prior to forecast April. 3
 GDP was the one determine electricity. — When we close the store on March 31 (6 pm), we N 16
18 + 28 + 30
= 𝟐𝟓. 𝟑
already know that we sold 13 cakes. With that, we 3
16 + 18 + 28
can now forecast for April—11.7. We already know D 14 = 𝟐𝟎. 𝟕
it before April 1, so we can prepare for our demand. 3
MOVING AVERAGES — We will know that we are able to sell 16 cakes
— In the 3-mth moving average, 106.3 is far from
For example, this happened in 2021. When we forecast during November is on November 30 (6 pm) when
 A moving average forecast uses a number of 350.
in 2022, never backtrack an abnormal year. If we think we close our store. So we will use this in order to
historical actual data values to generate a forecast. — Now, 69.6 is way too far from 350 because we used
this is how sales will be usually, then classify this as a forecast for December.
 Moving averages are useful if we can assume that a 5-mth moving average. 3 of those 5-month sale
normal year. — There will be instances that our sales in some
market demands will stay fairly steady over time. were so low.
months of the year will be increased.
 The most straightforward calculation, the average — In August, when we only used 3-month, we need to
Abnormal Year
price over a chosen time period. incorporate 200 (high), 100 (high), and 19 (low)—
— For example, a celebrity endorsed our cake on May.
which significantly affects the forecast that made it
Advantages: So sales on June increased.
very low.
(1) Moving averages can smooth out fluctuations in data, Month Actual 3-mth MA — If we will use 8-mth moving average, the forecast
making it easier to identify trends and patterns. J 10 will be too far as what happened in the past is
(2) Moving averages can be easily adjusted to reflect F 12 different from what is happening recently.
changes in the data, such as adding more periods to M 13 — As people knew that this would happen, they would
the moving average to capture longer-term trends or 13 + 12 + 10 use “Weighted Moving Averages.” They want to use
A 16 = 𝟏𝟏. 𝟕
decreasing the number of periods to capture short- 3 the numbers from the table but it is already way too
16 + 13 + 12 far.
term changes. M 19 = 𝟏𝟑. 𝟕
3
19 + 16 + 13
Disadvantages: J 100 = 𝟏𝟔
3 2-mth Moving Average
100 + 19 + 16
(1) Moving averages may not be accurate when dealing J 200 = 𝟒𝟓
with data that exhibits high volatility or seasonality, 3 Month Actual 2-mth MA
200 + 100 + 19 J 10
as they can fail to capture the full extent of these A 350 = 𝟏𝟎𝟔. 𝟑
changes.
3 F 12
350 + 200 + 100
(2) It is slower to respond to rapid price changes that S 28 = 𝟐𝟏𝟔. 𝟕 12 + 10
3 M 13 = 𝟏𝟏
often occur at market reversal points. 28 + 30 + 26 2
If our business is flowers, we should already be sourcing O 18 = 𝟐𝟖
3
for our February sales because of Valentine’s Day.

APRIL KAE APRIL KAE APRIL KAE APRIL KAE APRIL KAE APRIL KAE
2|Page
 13 + 12 Actual Forecast Errors |Error|  The weighted moving average is usually employed 3 (16) + 2 (18) + 1 (28)
A 16 = 𝟏𝟐. 𝟓 10 5 5 5 when there is a need to place more importance on D 14 6
2
15 20 -5 5 some periods over others. = 𝟏𝟖. 𝟕
16 + 13 20 15 5 5
M 19 = 𝟏𝟒. 𝟓  In most cases, we place more importance on more
2 25 30 -5 5 — (3, 2, 1) = 3 is the weight where we put the most
AVERAGE 0 MAD 5 recent data.
19 + 16  We give more weight on more recent observation. recent.
J 23 = 𝟏𝟕. 𝟓 — The denominator is 6 because it is the sum of the
2 — Forecast Errors = Actual – Forecast  WMA helps us highlight the importance of the more
recent data. weights (3 + 2 + 1 = 6).
23 + 19 — To see the average of the differences (Errors), add
J 26  The longer the series, the more it is irrelevant. — In April, the forecast when there is no weights yet
= 𝟐𝟏 them all and divide by the no. of errors.
2 is 11.7. While with the weights, it is 12.2
— “0” means that on the average, your forecast is the
26 + 23 — What really happened in April was 16 cakes, so the
A 30 = 𝟐𝟒. 𝟓 same as the actual sale (you have no mistakes).
2 Advantages: better forecast is with the weights, which is 12.2
However, looking at the sales and forecast, they are
because it is nearer to the actual sale.
30 + 26 not actually the same. (1) Weighted moving averages give more importance to
S 28 = 𝟐𝟖 — If the weights is (5, 2, 2), the denominator will be
2 — The errors each month were cancelled out so it did the most recent data points, which can help capture 9.
not depict what really happened. This tells us that short-term changes in trends and make the forecast
28 + 30
O 18 = 𝟐𝟗 getting the difference of actual and forecast is not more accurate. 5 (16) + 2 (13) + 5 (12)
2 = 𝟏𝟖. 𝟒
enough—so we must get the absolute value of each (2) The weights used in a weighted moving average can 9
18 + 28 error.
N 16 = 𝟐𝟑 be adjusted to reflect changes in the data, which can
2 — Get the average of the absolute values—Mean — What is closer to the actual sale is better.
help improve the accuracy of the forecast.
16 + 18 Absolute Deviation.
D 14 = 𝟏𝟕
2 — “5” means that on the average, we are 5 points Disadvantages:
4-mth Weighted Moving Average
away from what actually happened.
(1) Determining the weights to use in a weighted moving
Month Sales 4-mth MA
MEAN ABSOLUTE DEVIATION (MAD) average can be subjective and may require trial and
 We want a smaller MAD because it is more accurate (4, 3, 2, 1)
error.
and that it presents smaller amount of errors. J 10
 Mean Absolute Deviation is the sum of the absolute (2) Like traditional moving averages, weighted moving
 We want a smaller sum of the absolute forecast errors F 12
forecast errors over the number of errors. averages may not be as effective in situations where
(∑ |FE|) because it means that we are not far from the M 13
data has irregular patterns or is subject to sudden
∑ |FE| actual. A 16
changes.
 If the sum of the absolute forecast errors is smaller, 4 (16) + 3 (13) + 2 (12) + 1 (10)
# 𝑜𝑓 𝑒𝑟𝑟𝑜𝑟𝑠
then MAD will be smaller. M 19 10
= 𝟏𝟑. 𝟕
 Note that Error and Deviation are used 3-mth Weighted Moving Average
interchangeably. Compare! What is better: 3 MA or 2 MA? (YEAR 2022) 4 (19) + 3 (16) + 2 (13) + 1 (12)
Month Actual 3-mth MA J 23
 The two bars around “Forecast Error” denote absolute 10
(3, 2, 1) = 𝟏𝟔. 𝟐
values. That is, return a positive value whether the Month Actual 3 MA |FE| 2 MA |FE|
J 10
original value is positive or negative. J 10 4 (23) + 3 (19) + 2 (16) + 1 (13)
F 12
 To find the mean absolute deviation (MAD or MAE), F 12 J 26
M 13 10
we simply average these absolute errors—sum up the M 13 11 2 3 (13) + 2 (12) + 1 (10) = 𝟏𝟗.4
absolute errors, and then divide the total by sum of A 16 11.7 4.3 12.5 3.5 A 16 4 (26) + 3 (23) + 2 (19) + 1 (16)
6
errors. M 19 13.7 5.3 14.5 4.5 = 𝟏𝟐. 𝟐 A 30 10
J 23 16 7 17.5 5.5 3 (16) + 2 (13) + 1 (12) = 𝟐𝟐. 𝟕
Why deviation? J 26 19.3 6.7 21 5 M 19 6 4 (30) + 3 (26) + 2 (23) + 1 (19)
— Deviation is our error. A 30 22.7 7.3 24.5 5.5 = 𝟏𝟒. 𝟑 S 28 10
— How far our forecast is from our actual. S 28 26.3 1.7 28 0 3 (19) + 2 (16) + 1 (13) = 𝟐𝟔. 𝟑
Why absolute? O 18 28 10 29 11 J 23 6 4 (28) + 3 (30) + 2 (26) + 1 (23)
— Because we get the absolute value. N 16 25.3 9.3 23 7 = 𝟏𝟕 O 18 10
— If we don’t use absolute value, our errors will be D 14 20.7 6.7 17 3 3 (23) + 2 (19) + 1 (16) = 𝟐𝟕. 𝟕
cancelled. 58.3 / 9 47 / 10 J 26 6 4 (18) + 3 (28) + 2 (30) + 1 (26)
Why mean? MAD MAD = 𝟐𝟎. 𝟓
N 16 10
6.5 4.7 3 (26) + 2 (23) + 1 (19)
— Because we get the average. = 𝟐𝟒. 𝟐
A 30 6 4 (16) + 3 (18) + 2 (28) + 1 (30)
= 𝟐𝟑. 𝟖
D 14 10
How will you determine the better technique? CONCLUSION: Use a 2 MA for 2024. 3 (30) + 2 (26) + 1 (23)
= 𝟐𝟎. 𝟒
S 28 6
— If, for example, the forecasted figure is 16 and what — The better forecast is 2-month Moving Average = 𝟐𝟕. 𝟓
actually happen (sale) is 16, then that is the best. because of smaller MAD. It means that it is more 3 (28) + 2 (30) + 1 (26)
Compare! What is better: 3 WMA or 4 WMA? (YEAR 2022)
— If you forecast exactly or close to the actual sale, accurate and smaller mistakes. O 18 6
— The smallest MAD is what we will use next year. = 𝟐𝟖. 𝟑 Month Actual 3WMA |FE| 4 WMA |FE|
that is a better technique.
3 (18) + 2 (28) + 1 (30)
J 10
For example: N 16 6 F 12
WEIGHTED MOVING AVERAGES = 𝟐𝟑. 𝟑
M 13

APRIL KAE APRIL KAE APRIL KAE APRIL KAE APRIL KAE APRIL KAE
3|Page 4|Page
 A 16 12.2 3.8 = 0.1 Use the exponential smoothing to forecast August’s — The distance between the actual and the forecast
M 19 14.3 4.7 13.7 5.3 income. Assume that the initial forecast for February is (the red line) is the error—so the blue line is more
J 23 17 6 16.2 6.8 Quarter Actual Forecast $65,000. The smoothing constant selected is = 0.1. representative.
J 26 20.5 5.5 19.4 6.6 — In order to fix this problem, it is by calculus—to the
A 30 23.8 6.2 22.7 7.3 1 180 175 (assumed figure) Months Actual Forecast find the line that is nearest to the points.
S 28 27.5 0.5 26.3 1.7 — In MAD, we get the absolute value Σ |e| because
F 70,000 65,000 (assumed figure)
O 18 28.3 10.3 27.7 9.7 the errors are being cancelled out.
175 + 0.1 (180 – 175)
N 16 23.3 7.3 24.2 8.2 2 168 65,000 + 0.1 (70.000 – 65,000) — Similar to trend projection, we don’t want our errors
= 175.5 M 68,500
= 65,500
D 14 18.7 4.7 20.4 6.4 to be cancelled out.
49 / 9 52 / 8 65,500 + 0.1 (68,500 – 65,500) — The line gives us the minimum of the summation of
175.5 + 0.1 (168 – 175.5) A 64,800
MAD MAD 3 159 = 65,800
= 174.75 errors.
5.4 6.5 65,800 + 0.1 (64,800 – 65,800)
M 71,700
= 65,700
174.75 + 0.1 (159 – 174.75) 𝑚𝑖𝑛𝑖𝑚𝑖𝑧𝑒 𝛴 𝑒 2
4 175 65,700 + 0.1 (71,700 – 65,700)
= 173.18 J 71,300
CONCLUSION: Use a 3 WMA for 2024. = 66,300
173.18 + 0.1 (175 – 173.18) 66,300 + 0.1 (71,300 – 66,300) — When we squared error, it means that it becomes
5 190 J 72,800
= 173.36 = 66,800 positive (so the positive and the negative will not be
66,800 + 0.1 (72,800 – 66,800)
A cancelled out).
EXPONENTIAL SMOOTHING 173.36 + 0.1 (190 – 173.36) = 67,400
6 205 — The line that is closest to the point is Trend
= 175.02
 The time in forecasting is not necessarily months, it Projection Line.
could be days, weeks, quarters, or years. 175.02 + 0.1 (205 – 175.02) TREND PROJECTIONS o b = slope
7 180
 In moving average, for example, 5-month MA—the = 178.02 o a = constant
earlier month that we can forecast is June.  It is about finding the line that is nearest to the — We want a line that is closest to the points.
178.02+ 0.1 (180 – 178.02)
8 182 points.
= 178.22
𝐽+𝐹+𝑀+𝐴+𝑀  We can only use trend projection if we think the
5 variable will continue with that slope two-time.
— 175 (forecast) and 180 (actual) will be used to
o If we are considering a variable that will
forecast 2nd Quarter.
 This is the disadvantage of moving average that eventually stagnate, instead of continually, we
— In order to know the actual sales for the second
even if we have a lot of sales in April compared to can never use trend projection.
quarter, it is determined on June 30 at 6pm (end of
the others, we are still hold back by months with o We can only use trend projection to project
2nd quarter)—168.
low sales—so exponential smoothing solve that outside the range (if we stop in 2022, we can
— In order to know the actual sales for the third
problem. only project for 2024).
quarter, it is determined on September 30 at 6pm
 Exponential Smoothing only goes by 1 period. o The assumption is the line continuous in that
(end of 3rd quarter)—159.
 They use weighted averages of past observations to same slope for the years you will use it when
— We will know that our forecast is far from the actual
forecast new values. we project.
sale in the 4th quarter during December 31 at 6pm.
 Exponential smoothing is generally used to make  Trend projection uses your past sales data to
short term forecasts, but longer-term forecasts 1st Quarter J+F+M project your future sales.
using this technique can be quite unreliable. 2nd Quarter A+M+J  This method requires a long time-series data.
𝑻𝒓𝒆𝒏𝒅 𝑷𝒓𝒐𝒋𝒆𝒄𝒕𝒊𝒐𝒏 𝑳𝒊𝒏𝒆 → 𝑦̂ = 𝑎 + 𝑏𝑥
3rd Quarter J+A+S  This involves analyzing historical data to identify
New Forecast = last period’s forecast + (last period’s 4th Quarter O+N+D trends and using those trends to make predictions
o y-hat because we represent the actual sales by y.
actual sales – last period’s forecast) about future values.
o But when the actual sales were represented in the
— Note that we cannot start forecasting using point of a line. Those points in a line are just an
= smoothing constant exponential smoothing without 1 data in forecast estimation.
column. o Above the line = positive error
— If the data we are currently forecasting is for the o Below the line = negative error
Advantages: year 2023, in order to get that 1 forecast, we can o x = y (actual sale)
get the last forecast of 4th quarter of year 2022. o points in a line = y-hat (forecast/estimates)
(1) Exponential smoothing is best used for forecasts that
— Another way to know the start our forecast for the
are short-term and in the absence of seasonal or
year 2023 is the forecast of the 1st quarter of 2022, 𝑛 𝛴 𝑥𝑦 − 𝛴𝑥𝛴𝑦
cyclical variations. 𝒃=
as it seems like same occurrences. 𝑛 𝛴 𝑥 2 − (𝛴𝑥 )2
(2) It produces accurate forecasts as it produces a
forecast for one period ahead.
2nd EXAMPLE: 𝒂 = 𝑦̅ − 𝑏𝑥̅
Disadvantages:

(1) Forecasts aren’t accurate when data with cyclical or = 0.1 𝒏 = 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑦𝑒𝑎𝑟𝑠 𝑡𝑜 𝑓𝑜𝑟𝑒𝑐𝑎𝑠𝑡
seasonal variations are present. — We want the blue line because it is the best—it
Consulting income at Kate Walsh Associates for the means that the projection is close to the actual
period February – July has been follows: sales.
1ST EXAMPLE:
— We want our Actual minus Forecast be smaller or
closer. X New X Y
1ST EXAMPLE:

APRIL KAE APRIL KAE APRIL KAE APRIL KAE APRIL KAE APRIL KAE
5|Page 6|Page
Year Code Demand x2 xy 2017 3 — They have a positive relationship or directly
2015 1 74 1 74 2018 4 proportionate to each other because they need the
2019 5 Trend Projection Line same thing to grow (i.e., both of them are
2016 2 79 4 158
2020 6 dependent of rain, sun, and soil) but it doesn’t
2017 3 80 9 240 𝑦̂ = 𝑎 + 𝑏𝑥
2021 7 cause each other.
2018 4 90 16 360 2022 8 ̂ = 66,920.2 + (837.1)𝑥 — We can use regression analysis to find out their
𝒚
2019 5 105 25 525 2023 9 relationship. But it’s not a causal relationship.
2020 6 142 36 852 2024 10 X New X — Regression Analysis under Causal Forecasting is
Year Code actually misnomer because it is not all the time that
2021 7 122 49 854
F 1 when we use regression analysis, one causes the
𝜮𝒙 𝜮𝒚 𝜮 𝒙𝟐 𝜮 𝒙𝒚 Forecast for 2024 other. It is only when the body of knowledge that
M 2
= 28 = 692 = 140 = 3,063 supports the theory.
̂𝟐𝟎𝟐𝟒 = 56.9 + 10.5 (10)
𝒚 A 3
𝒙̅ 𝒚̅ — But we can use variable that are directly
M 4 proportional to each other.
=4 = 98.9 ̂𝟐𝟎𝟐𝟒 = 161.9
𝒚
J 5
— 2015 x 74 or 1026 x 79 is a big number. So what J 6
we can do so that our computation is not so tedious — The slope (b) can be negative. There can be line A 7
is to code—we can just say 1, 2, 3, 4, 5, 6, 7 (coding going down.
years)
— x2 = squaring the new x (codes) Forecast for August
2nd EXAMPLE:
— xy = multiplying x (new x) by y (demand)
̂𝑨𝒖𝒈𝒖𝒔𝒕 = 66,920.2 + (837.1) (7)
𝒚
Consulting income at Kate Walsh Associates for the
𝑛 𝛴 𝑥𝑦 − 𝛴𝑥𝛴𝑦 period February – July has been follows: ̂𝑨𝒖𝒈𝒖𝒔𝒕 = 72,779.9
𝒚
𝒃= 2
𝑛 𝛴 𝑥 2 − (𝛴𝑥) Use the trend projections and forecast for August. Example:

7 (3,063) − (28) (692) X New X Y — The more money we put in advertisement, the more
𝒃= REGRESSION ANALYSIS
2 Month Code Income x2 xy sales will be.
7 (140) − (28)
F 1 70,000 1 70,000  This is under Causal Forecasting — Increase in advertisement cause increase in sales—
2,065  There is a y and an x. we have body of knowledge to say that.
𝒃= M 2 68,500 4 137,000
196  y = f (x) — This is cause and effect.
A 3 64,800 9 194,400
o when y is a function of x
𝒃 = 10.5 M 4 71,700 16 286,800
o x really causes y
J 5 71,300 25 356,500  We can use regression analysis to find out the
— The difference between 𝛴 𝑥 2 and (𝛴𝑥 )2 is that (𝛴𝑥 )2
J 6 72,800 36 436,800 relationship of the variable. But it’s not a causal
is about getting the sum of x and then squared—
relationship (one does not cause the other).
282 𝜮𝒙 𝜮𝒚 𝜮 𝒙𝟐 𝜮 𝒙𝒚
 Regression Analysis is not always cause and effect,
— While 𝛴 𝑥 2 is about squaring each x and add them = 21 = 419,100 = 91 = 1,481,500
 It can only be cause and effect if there is a body of
all. 𝒙̅ 𝒚̅ knowledge that supports that theory.
= 3.5 = 69,850  There are variables that can grow together even if
𝒂 = 𝑦̅ − 𝑏𝑥̅ one does not cause the other.
𝑛 𝛴 𝑥𝑦 − 𝛴𝑥𝛴𝑦 Example: Example:
𝒂 = 98.9 − (10.5)(4) 𝒃=
𝑛 𝛴 𝑥 2 − (𝛴𝑥)2
— x = income — Historical data (CQPA) of those who graduated from
𝒂 = 56.9 6 (1,481,500) − (21) (419,100) — y = sale of luxury of goods Miriam College.
𝒃= — As income increases, sale of luxury goods increases
6 (91) − (21)
2 — All CQPAs in every IQ follows a normal curve.
— We cannot say that an increase in income CAUSES — If we will attach all means of those normal curve,
Trend Projection Line 87,900 the increase in sale of luxury goods. we will be able to form regression line (y = a + bx)—
𝒃= — This is because there is a body of knowledge that y-hat.
105
𝑦̂ = 𝑎 + 𝑏𝑥 support we say. — e.g., We want to forecast Mayumi’s CQPA 3 years
𝒃 = 837.1 — We can only say that x causes y if we are supported from now given her IQ of 115.
̂ = 56.9 + 10.5𝑥
𝒚 by literature. — The forecast is that Mayumi will have a CQPA of
— If product 1 increases and product 2 increases but almost perfect 5 when she graduates from now. But
— If we want to project for 2024, the “x” we will
nothing supports us we cannot say that product 1 it is not guaranteed that she will really graduate
substitute above is 10. 𝒂 = 𝑦̅ − 𝑏𝑥̅ causes product 2. with an average of 6, since it is a normal curve, all
X New X 𝒂 = 69,850 − (837.1)(3.5) values there are possible. She can even have an
Example:
Year Code average of 3.
2015 1 𝒂 = 66,920.2 — Length of grass dos not cause height of trees.
2016 2

APRIL KAE APRIL KAE APRIL KAE APRIL KAE APRIL KAE APRIL KAE
7|Page 8|Page
 — 5 is what the line projects. It is more probable that X Y 2nd EXAMPLE: 𝜮𝒙 𝜮𝒚 𝜮 𝒙𝟐 𝜮 𝒙𝒚
she will get a 5 because it is where the normal curve 1st Grade Final Grade x2 xy = 132 = 271 = 1,796 = 3,529
ups. Bus and subway ridership in Washington, D.C., during the
98 93 9,604 9,114 𝒙̅ 𝒚̅
summer months is believed to be heavily tied to the
77 78 5,929 6,006 = 11 = 22.6
number of tourists visiting the city. During the past 12
Difference between Trend Projections and Regression 88 84 7,744 7,392 years, the following data have been obtained:
Analysis 80 73 6,400 5,840 𝑛 𝛴 𝑥𝑦 − 𝛴𝑥𝛴𝑦
(a) Plot these data and decide if a linear model is 𝒃=
TREND PROJECTION 96 84 9,216 8,064 𝑛 𝛴 𝑥 2 − (𝛴𝑥)2
reasonable.
61 64 3,721 3,904 (b) Develop a regression relationship.
— Time against any variable. 12 (3,529) − (132) (271)
66 64 4,356 4,224 (c) What is expected ridership if 10 million tourists visit 𝒃= 2
the city? 12 (1,796) − (132)
95 95 9,025 9,025
(d) If there are no tourists at all, explain the predicted
69 76 4,761 5,244 6.576
ridership. 𝒃=
𝜮𝒙 𝜮𝒚 𝜮 𝒙𝟐 𝜮 𝒙𝒚 4,128
= 730 = 711 = 60,756 = 58,813 (a) Plot these data and decide if a linear model is
𝒃 = 1.6
𝒙̅ 𝒚̅ reasonable.
= 81.1 = 79 0 = 5 + 1.6𝑥 𝑦̂ = 5 + 1.6 (0)
−5 = 1.6𝑥
̂ = 𝟓 + 𝟏. 𝟔𝒙 1.6𝑥
𝒂 = 𝑦̅ − 𝑏𝑥̅
𝒚
𝑛 𝛴 𝑥𝑦 − 𝛴𝑥𝛴𝑦
REGRESSION ANSLYSIS 𝒃= 2 𝒂 = 22.6 − (1.6)(11)
𝑛 𝛴 𝑥 2 − (𝛴𝑥) 𝒙 = (−𝟑. 𝟏𝟐𝟓, 𝟎) 𝒚 = (𝟎, 𝟓)
— Any x and y. 𝒂= 5
9 (58,813) − (730) (711)
— Two variables that one not necessarily cause the 𝒃= 2
other but when there is a body of knowledge that 9 (60,756) − (730)
says that one causes another, then x causes y.
10,287 Regression Line
𝒃=
13,904
𝑦̂ = 𝑎 + 𝑏𝑥
𝒃 = 0.7
̂ = 5 + 1.6𝑥
𝒚
— Positive 0.7 = it means that the 2 variables are
positively related.
(c) What is expected ridership if 10 million tourists visit
the city?
1ST EXAMPLE: 𝒂 = 𝑦̅ − 𝑏𝑥̅ Expected Ridership
Students in a Management Science class just have 𝒂 = 79 − (0.7)(81.1) A linear model is reasonable because as the number of ̂𝑹𝒊𝒅𝒆𝒓𝒔𝒉𝒊𝒑 = 5 + 1.6 (10)
𝒚
received their grades on the first test. The instructor has
tourists increases, the ridership appears to increase.
provided information about the first test grades in some 𝒂 = 22.2
previous classes as week as the final average for the
̂𝑹𝒊𝒅𝒆𝒓𝒔𝒉𝒊𝒑 = 2,100,000 𝑟𝑖𝑑𝑒𝑟𝑠
𝒚
(b) Develop a regression relationship.
same students and are as follows:
X Y Do not put 10,000,000 in x, instead only 10. Because we
Student 1 2 3 4 5 6 7 8 9 Regression Line Number of initially did not apply the millions in the table.
Ridership
Year Tourists x2
xy
1st Grade 98 77 88 80 96 61 66 95 69 (100,000’s)
𝑦̂ = 𝑎 + 𝑏𝑥 (1,000,000’s)
1 7 15 49 105 (d) If there are no tourists at all, explain the predicted
Final
93 78 84 73 84 64 64 95 76 ̂ = 22.2 + 0.7𝑥
𝒚 ridership.
Average 2 2 10 4 20
3 6 13 36 78 ̂ = 5 + 1.6 (0)
𝒚
Alyssa got 83 on her 1st grade. What could be her final 4 4 15 16 60 ̂=5
𝒚
average? Predict the final average of Alyssa, who got an x = 83 (1st grade Alyssa got)
5 14 25 196 350
83 on her 1st Test. Final Average (expected to be) If there are no tourists at all, the model predicts a
6 15 27 225 405
ridership of 5 or 500,000 persons.
Let the first grade be x and the final average be y. 7 16 24 256 384
̂𝑭𝒊𝒏𝒂𝒍 𝑨𝒗𝒆𝒓𝒂𝒈𝒆 = 22.2 + 0.7 (83)
𝒚
8 12 20 144 240
— Not necessarily that the first grades will have an Note:
affect on the final grade/average, but it may be an ̂𝑭𝒊𝒏𝒂𝒍 𝑨𝒗𝒆𝒓𝒂𝒈𝒆 = 80.3
𝒚 9 14 27 196 378
o This is causal relationship. The tourist cause an
indicator. 10 20 44 400 880 increase in ridership.
— But if the students do well in the first test, it is — What we get in our model are averages. Alyssa can
11 15 34 225 510 o x (independent variable) and y (variable that
probably that their final grade is high. even get 89, 77, or others. But on the average, we
12 7 17 49 119 depends on x)
expect Alyssa to get a final average of 80.

APRIL KAE APRIL KAE APRIL KAE APRIL KAE APRIL KAE APRIL KAE
9|Page 10 | P a g e