BUSINESS STATISTICS

ASSIGNMENT

SUBMITTED TO:

PROF. MIDDELA MOUNISAI SIDDHARTHA

SUBMITTED BY:

VANSHIKA GUPTA - 2501224

TISHA - 2501221

KARTIKEY KASHYAP - 2501189

YUVRAJ BHAMA - 2501229

MOHIT SAHOO - 2501195

PRATYUSH - 2501202
Rationale for Data Cleaning: Data cleaning was undertaken to eliminate physiologically impossible
values, correct measurement errors, and refine extreme outliers. This ensured the dataset reflected
plausible health characteristics of the target population. By improving accuracy and consistency, the
cleaned data provided a reliable foundation for valid statistical analysis and meaningful interpretation.

Statistics in Health Data Interpretation

For the purpose of this practical exercise, our dataset is treated as the complete population. This
approach allows for a direct comparison between population parameters and sample statistics, providing
a clear and tangible demonstration of each concept.

Strategic Sampling Methods for Robust Data Collection

A key initial step in any statistical study is obtaining a representative sample. If a sample is systematically
unrepresentative, it can lead to significant biases, like selection bias or nonresponse bias, which
invalidate the statistical inferences drawn from the data. One technique for ensuring representativeness is
stratified random sampling. This involves dividing the population into mutually exclusive and collectively
exhaustive subgroups, known as strata, and then drawing a random sample from each stratum.

Application: Stratified Random Sample based on Gender

To illustrate this, a stratified random sample was drawn from the heart patient dataset, with the gender
variable serving as the stratification criterion. The population was first divided into two strata:

gender=1 (female) and gender=2 (male). The initial population distribution was determined to be
approximately 65% female and 35% male. To ensure the sample mirrored this distribution, a stratified
sample of 50 patients was selected, with 19 males (62%) and 31 females (38%).

This method is especially valuable in clinical research, where risk factors and outcomes for conditions like
cardiovascular disease can differ between males and females. A simple random sample might over- or
under-represent one gender, leading to biased conclusions. Stratified sampling mitigates this risk, ensuring
gender-specific trends are accurately captured in the data. The choice of sampling method, therefore,
represents a fundamental trade-off between the increased precision and reliability of methods like stratified
sampling and the logistical challenges and costs associated with them.

Gender Category Population Count Population Sample Count Sample Percentage

(Population) Percentage (Sample)

Female (1) 23891 34.87% 19 38%

Male (2) 44615 65.13% 31 62%

Total 68506 100% 50 100%

The Sampling Distribution
While a population parameter is a fixed, unknown constant, a sample statistic is a random variable whose
value depends on the specific sample selected. By understanding the distribution of all possible sample
means, it becomes possible to quantify the uncertainty surrounding a single sample's estimate. The central
principle here is that averaging across a sample reduces the extreme values and therefore reduces
variability. The standard deviation of this sampling distribution is known as the standard error, and it is a
measure of the sample statistic's variability from sample to sample.

Application: Sampling Distribution for Body Mass Index (BMI)

To make this abstract concept concrete, a simulation was performed using the BMI variable from heart
patient dataset. This variable, calculated from a patient's height & weight, is a continuous measure of body
fat. The population mean BMI was calculated to be approximately 27.44, with a standard deviation of 5.17.

To generate the sampling distribution of the mean BMI, a process was simulated where 1,000 independent
random samples, each of size n=50, were drawn from the population. The mean BMI was calculated for
each of these 1,000 samples. The frequency distribution of these 1,000 sample means constitutes the
sampling distribution. (Sample 1000 sheet)

This confirms the assertion that averaging reduces variability. In a clinical context, this demonstrates why a
large number of independent patient BMI readings are a more stable and trustworthy indicator of population
health than any single patient's BMI.

The Central Limit Theorem

The Central Limit Theorem (CLT) posits that for a sufficiently large sample size, typically considered to be
30 or more observations, the sampling distribution of the mean will be approximately normally distributed,
regardless of the shape of the original population distribution.

Application: Demonstrating Normality with BMI Sample Means

This is of immense practical importance in health research. In many real-world scenarios, it is impossible to
know the exact distribution of a variable in the entire population. The CLT provides the theoretical
justification for proceeding with standard statistical tests and confidence interval calculations on the
sample data, as long as the sample size is large enough. This concept together with the law of large
numbers, ensures the sample mean is a reliable estimate with measurable uncertainty.

Point and Interval Estimates

Inference is the process of using sample data to make statements about a population. There are two
primary types of estimates used for this purpose: point estimates and interval estimates.

A point estimate is a single, numerical value calculated from a sample, which serves as the "best guess"
for population parameter. Example, the sample mean (x̄) is a point estimate for the population mean μ.

An interval estimate, also known as a confidence interval, addresses this limitation by providing a range of
values within which the true population parameter is likely to be found. This interval provides a measure of
the precision of the estimate and the level of certainty associated with it.

Application: Point Estimate for Systolic Blood Pressure (ap_hi)

To demonstrate this, a single random sample of 50 patients was drawn from the dataset. The variable of
interest is ap_hi, which represents a patient's systolic blood pressure. The population mean systolic blood
pressure was found to be approximately 126.6, with a standard deviation of 16.5. (Base Data). From the
sample of 50 patients, the mean systolic blood pressure was calculated to be 125.5. This is our point
estimate for the true mean systolic blood pressure of the entire patient population. The sample standard
deviation was 14.0. (Sample 50 sheet)
Reporting only this point estimate would be incomplete, as it fails to convey the reliability of the number.
The sample mean of 125.8, while our best single estimate, could be quite different from the true population
mean. A sample of 50 is small relative to the population, and another sample of 50 would almost certainly
yield a different mean. The interval estimate is the next logical step, providing a framework for
understanding how close our point estimate is to the truth.

Constructing a Confidence Interval for the Mean (Z/T)

A confidence interval provides a range of plausible values for an unknown population parameter, based on
a sample statistic. The confidence level, chosen by the researcher, indicates the percentage of intervals
that would capture the true population mean if the sampling method were repeated many times. The
formula for a confidence interval for the mean is:

Confidence Interval = Sample Mean ± Margin of Error, where the Margin of Error is calculated as:

Margin of Error = Critical Value × Standard Error, Standard Error = 𝒔/√𝒏

The choice between a critical Z-value and a T-value depends on the sample size and whether the
population standard deviation is known. Given a large sample size of 50 (n≥30), the Z-distribution can be
used. For a 95% confidence level, the critical Z-value is 1.96.

Application: 95% Confidence Interval for Mean Systolic Blood Pressure (ap_hi)

Using the sample data from the previous section (n=50, x̄ =125.5, s=14.0), a 95% confidence interval for
the mean systolic blood pressure of the population was constructed.
1. Calculate the Standard Error (SE): SE = 𝑠/√𝑛 = 14/ √50 = 1.98
2. Calculate the Margin of Error (ME): ME = Z × SE = 1.96×1.98 = 3.88
3. Construct the Confidence Interval: Confidence Interval = x̄ ± ME=125.5 ± 3.88 leading to Lower Bound
= 121.62, Upper Bound = 129.38

Application: 99% Confidence Interval for Mean Systolic Blood Pressure (ap_hi)

Using the sample data from the previous section (n=50, x̄ =125.5, s=14.0), a 99% confidence interval for
the mean systolic blood pressure of the population was constructed.

1. Calculate the Standard Error (SE): SE = 𝑠/√𝑛 = 14/ √50 = 1.98

2. Calculate the Margin of Error (ME): ME = Z × SE = 2.57×1.98 = 5.088
3. Construct the Confidence Interval: Confidence Interval = x̄ ± ME=125.5 ± 5.088 with Lower Bound =
120.412, Upper Bound = 130.588

The 95% confidence interval for the mean systolic blood pressure is approximately [121.62, 129.38]. The
interpretation is that if this method of creating a confidence interval were repeated with many different
samples, approx. 95% of those intervals would contain the true population mean. A crucial relationship
exists between the margin of error, the confidence level, & sample size. To achieve a higher level of
confidence (e.g., 99%), a larger critical Z-value is required, which in turn widens the interval and
decreases the precision of the estimate.

Constructing a Confidence Interval for a Proportion

Proportions are used when the variable of interest has a binary outcome, such as the presence or
absence of a disease. The formula for a confidence interval for a proportion is similar in structure to that
for the mean, but the calculation of the standard error is different:

Confidence Interval = Sample Proportion (p̂) ± Margin of Error, where the Margin of Error is:

Margin of Error = Critical Z-value × Standard Error

Standard Error = √𝒑̂(𝟏 − 𝒑̂)/𝒏

Application: 95% Confidence Interval for Cardiovascular Disease (cardio)
To demonstrate this, the same random sample of 50 patients was used, but this time the cardio variable
was analyzed.2 This variable indicates the presence (1) or absence (0) of cardiovascular disease. The
sample was found to have 31 patients with cardiovascular disease. (Sample of 50)
1. Calculate the Sample Proportion (p̂): p̂= Number of cases/Sample size = 31/50 = 0.62
2. Calculate the Standard Error (SE): SE= √𝑝̂(1 − 𝑝̂)/𝑛 = √0.62(1 − 0.62) = 0.0686
3. Calculate the Margin of Error (ME): Using the same 95% confidence level, the critical Z-value remains
1.96. ME=Z×SE=1.96 × 0.0686 ≈ 0.1345
4. Construct the Confidence Interval: Confidence Interval = p̂ ± ME=0.62 ± 0.1345, Lower Bound = 0.62
− 0.1345 = 0.4855 and Upper Bound = 0.62 + 0.1345 = 0.7545

The 95% confidence interval for the proportion of patients with cardiovascular disease is approx. [0.4855,
0.7545]. This means that, based on our sample, we are 95% confident that the true percentage of
patients with cardiovascular disease in the entire population lies between 48.55% and 75.44%.

Determining the Optimal Sample Size

A sample size that is too small may result in a study that is underpowered and unable to detect significant
effects, rendering the findings inconclusive. Conversely, a sample that is unnecessarily large wastes
valuable time and resources. The process of sample size determination involves a careful consideration of
three key criteria:

1. The desired level of confidence,

2. The acceptable margin of error (or precision), and
3. The degree of variability or variance in the population being studied.

The relationship between these criteria is often formalized through a calculation for the required sample
size, such as the following formula for a proportion: n= (z ɑ/2 𝛔 /E)2 or n = ((z ɑ/2 /E)2 * p̂(1-p̂)

● n is the required sample size.

● Z ɑ/2 is the critical Z-value for the desired confidence level.
● 𝜎 is the estimated population standard deviation
● E is the desired margin of error (precision).
● p̂ is represents the sample proportion.

Application: Sample Size for Smoking Prevalence

Imagine a scenario where we are planning a new study and want to estimate the proportion of heart
patients who are smokers (smoke=1). We do not have any prior information from a pilot study. Our
research objective is to be 95% confident that our estimate is within a margin of error of ±5%.

1. Identify the parameters:

○ Confidence Level: 95%, so the critical Z-value is 1.96.

○ Desired Precision (Margin of Error, E): 0.05 (or 5%).
○ Population Variability (p): We will take the variability of smokers as
p(1−p), from the population (Base Data) we see that p̂ = 0.097
2. Calculate the required sample size: n = (1.96)2 x 0.1x(1-0.1)/(0.05)2 = (3.8416 x 0.09)/0.0025 = 138

n = The required sample size is approximately 138 patients. This demonstrates a relationship
between the chose nmargin of error and the resources needed. For instance, if a researcher wanted to
increase the precision to a margin of error of ±3%, the required sample size would increase to 384
patients. This illustrates that sample size is a direct and calculable consequence of the study's
objectives regarding confidence and precision, tempered by the expected variability of the population.
Fundamentals of Hypothesis Testing

Calculating Population Parameters

Before conducting any hypothesis tests on samples, it is essential to establish the true parameters of our
defined population. These parameters represent the "ground truth" that our null hypotheses will be based
on. The following parameters have been calculated.

Parameter Description Population Value

Population Mean Body Mass Index (BMI) 27.44
Population Standard Deviation of BMI 5.18
Population Mean Systolic Blood Pressure 126.68
Population Proportion of Smokers 0.09

These values are the benchmarks for our analyses. For instance, when we test a sample's mean BMI, our
null hypothesis will state that the mean is equal to the population value of 27.44

z-Test for Mean (Analysis of Body Mass Index - BMI)

Scenario and Problem Statement: A health analytics firm wants to assess the impact of a public health
intervention on BMI. The known population mean BMI is 27.44 (σ = 5.18). They collect a random sample of
50 individuals to test if the current average BMI differs from the population mean. Since the population
standard deviation is known and the sample size is large, a z-test is appropriate.

Hypotheses

• H0:μ=27.44 (no change)

• H1:μ≠27.44 (significant change)

Significance Level: α=0.5, this means we are willing to accept a 5% risk of concluding there is a difference
when one does not actually exist (a Type I error).

Sample BMI Data (n=50): 24.3, 27.3, 26.9, 23.9, 25.4, 28.7, 26.5, 23.5, 30.8, 30.1, 33.5, 22.1, 25.1, 29.8,
26.8, 25.2, 29.4, 25.1, 22.1, 23.5, 30.1, 25.8, 23.5, 26.5, 28.7, 25.4, 23.9, 26.9, 27.3, 24.3, 24.3, 27.3, 26.9,
23.9, 25.4, 28.7, 26.5, 23.5, 25.8, 30.1, 23.5, 22.1, 25.1, 29.8, 26.8, 25.2, 29.4, 25.1, 22.1, 23.5

• Sample Size (n): 50

• Sample Mean (x̄): 26.15

Decision via Critical Value Method

Test Statistic:
𝑥̅ − 𝜇
𝑧 = 𝜎
√𝑛
26.15 − 27.44 −1.29
𝑧 = = ≈ − 1.762
0.732 0.732

where x̄ is the sample mean, μ is the hypothesized population mean, σ is the population standard deviation,
and n is the sample size.

• Critical value = zcritical = ±1.96.

• Since ∣−1.762∣=1.762, which is not greater than 1.96, our test statistic does not fall into the rejection
region. We fail to reject the null hypothesis.
Decision via P-Value Method

• Calculate P-Value: The probability of getting a z-score of -1.762 or less is P(Z≤−1.762). Using Excel's
[Link](-1.762, TRUE) function, this is approximately 0.039. Since this is a two-tailed test, we
multiply this probability by 2.
𝑝̂-value = 2 × 𝑃(𝑍 ≤ −1.762) = 2 × 0.039 = 0.078
• Since 0.078 is greater than our significance level of 0.05, We fail to reject the null hypothesis.

Interpretation of Findings

Both the critical value (|−1.762| < 1.96) and p-value (0.078 > 0.05) methods lead to failing to reject the null
hypothesis. At a 5% significance level, there’s insufficient evidence to say the average BMI changed from
27.44. The observed sample mean (26.15) likely reflects random variation, so the health intervention hasn’t
shown a statistically significant impact.

T-Test for Mean (Analysis of Systolic Blood Pressure - ap_hi)

Scenario and Problem Statement: A clinician suspects the average systolic blood pressure in a
community differs from the healthy level of 120 mmHg. Since the population standard deviation is unknown,
a random sample of 50 patients is collected. A one-sample t-test is appropriate to compare the sample
mean against the hypothesized value, using the sample standard deviation for estimation.

Hypotheses: H0:μ=120 ; HA:μ≠120

Significance Level = 0.05

Sample ap_hi Data (n=50): 120, 110, 140, 120, 120, 130, 120, 140, 110, 130, 140, 120, 110, 130, 120,
140, 120, 120, 110, 140, 120, 130, 120, 140, 110, 130, 140, 120, 110, 130, 120, 140, 120, 120, 110, 140,
120, 130, 120, 140, 110, 130, 140, 120, 110, 130, 120, 140, 120, 120

• Sample Mean (x̄): 124.8

• Sample Standard Deviation (s): 10.54

Decision via Critical Value Method

x̅−μ
t-Statistic: t = s
√n
124.8 − 120 4.8
t= = ≈ 3.221
1.49 1.49

Degrees of Freedom (df): df = 50 − 1 = 49

α=0.05 & df=49, critical t-value = [Link].2T(0.05, 49) function, we get 𝒕critical ≈ ±𝟐. 𝟎𝟏

• T-statistic = ∣3.221∣ = 3.221 > 2.01, our test statistic falls into the rejection region.
• Conclusion: We reject the null hypothesis.

Decision via P-Value Method

• P-Value: For a two-tailed test, we use Excel's [Link].2T(3.221,49) function. p−value ≈ 0.002 < 0.05.
• Conclusion: We reject the null hypothesis.
Interpretation: Both methods lead to rejecting the null hypothesis. The p-value of 0.002 shows a 0.2%
chance of observing a sample mean of 124.8 mmHg if the true average were 120 mmHg. This provides
strong evidence that the community’s average systolic blood pressure is significantly higher than the
recommended 120 mmHg.

z-Test for Proportion (Analysis of Smoking Rate)

Scenario and Problem Statement

A national health organization reports an 8.8% adult smoking rate. A local health department surveys 50
residents to see if their community’s rate differs from this benchmark. A one-proportion z-test is used to
compare the sample proportion against the known population proportion.

Hypotheses

• H0:P=0.088
• Ha:P≠0.088 (two-tailed test)

Significance Level: α=0.05

Sample Data

A random sample of 50 residents was surveyed. The results are as follows:

• Sample Size (n): 50

• Number of Smokers (Successes, x): 7
x 7
• Sample Proportion (p̂): = = 0.14
n 50

Assumption Check:

𝑛 × 𝑃 = 50 × 0.088 = 4.4

𝑛 × (1 − 𝑃) = 50 × (1 − 0.088) = 50 × 0.912 = 45.6

where p̂ is the sample proportion, P is the hypothesized population proportion, and n is the sample size.

Decision via Critical Value Method

̂ −P
p
Z-Statistic: z =
P(1−P)
√
n

0.14 − 0.088 0.052

z= = ≈ 1.297
0.0401 0.0401

• Critical Values: For a two-tailed test with α=0.05, the critical z-values are ±1.96.
• Decision Rule: Reject H0 if ∣ zcalculated ∣>1.96.
• Since ∣1.297∣<1.96, the test statistic does not fall into the rejection region.
• Conclusion: We fail to reject the null hypothesis.

Decision via P-Value Method

• Calculate P-Value: P(Z > 1.297) = 1 − P(Z ≤ 1.297)

Using Excel's 1 - [Link](1.297, TRUE), this is 1−0.9027=0.0973
p − value = 2 × 0.0973 = 0.1946
• Decision Rule: Reject H0 if p-value ≤α.
• The p-value of 0.1946 is greater than α=0.05.
• Conclusion: We fail to reject the null hypothesis.

Interpretation:

The sample smoking rate (14%) exceeds the national average (8.8%) but isn’t statistically significant (p-
value = 0.1946) with a sample size of 50. The local health department lacks sufficient evidence to conclude
a difference; a larger sample is needed.
Comparing means of two Independent Variables

Hypothesis:

H0: Μ1≥Μ2
HA: Μ1<Μ2

Μ1=Mean BP for Weight <= 100

Μ2=Mean BP for Weight >100

t-Test: Two-Sample Assuming Unequal Variances

Blood Pressure when weight < 100 Blood Pressure when weight > 100
Mean 126.58 139.19
Variance 259.06 205.07
Observations 179.00 75.00
Hypothesized Mean
Difference 0.00
df 155.00
t Stat -6.17
P(T<=t) one-tail 0.00
t Critical one-tail 1.65
P(T<=t) two-tail 0.00
t Critical two-tail 1.98

The results show a statistically significant difference in mean blood pressure between people weighing less
than 100 and those weighing more than 100.

Key Results:

Mean blood pressure for weight < 100: 126.58

Mean blood pressure for weight > 100: 139.19

t Stat: -6.17

p-value (one-tail): 0.00

t Critical (one-tail): 1.65

Inference:

• The p-value is much less than 0.05 (commonly used significance level), and the t-Stat is far below the
critical value. This means we reject the null hypothesis.
• There is strong evidence that people weighing less than 100 have significantly lower mean blood
pressure than those weighing over 100.
• The test results support your alternative hypothesis: high weight is associated with higher blood
pressure in the data.

In plain language: People who weigh above 100 have higher average blood pressure, and this difference is
statistically meaningful.

F-test for ratio of variances

F-Test Two-Sample for Variances

Blood Pressure when weight < 100 Blood Pressure when weight >100
Mean 126.58 139.19
Variance 259.06 205.07
Observations 179.00 75.00
df 178.00 74.00
F 1.26
P(F<=f) one-tail 0.13
F Critical one-tail 1.40

The F-test results show that the F value is 1.26 and the p-value is 0.13, which is higher than the standard
threshold of 0.05.

Inference:

• Since the p-value (0.13) >0.05, we do not have enough evidence to say the variances (spread) of blood
pressure are different between people with weight less than 100 and those with weight more than 100.
• The variability in blood pressure readings is similar for both low weight and high weight groups.
Therefore, there is no significant difference in how spread out the blood pressure values is for the two
weight categories.
Analysis of Variance (ANOVA)
The objective of this section is to apply Analysis of Variance (ANOVA) techniques to examine whether
certain health indicators differ significantly across patient groups. Specifically, a One-Way ANOVA and a
Two-Way ANOVA are performed to analyse variations in Body Mass Index (BMI) across selected
categorical health factors.

One-Way Anova: The purpose of the One-Way ANOVA is to test whether BMI differs significantly across
different cholesterol categories.

• Null Hypothesis (H₀): Mean BMI is equal across all cholesterol groups.
• Alternative Hypothesis (H₁): At least one cholesterol group has a different mean BMI.

Variables Used

• Dependent Variable: BMI (continuous).

• Independent Variable: Cholesterol (categorical: 1 = normal, 2 = above normal, 3 = well above normal).

SUMMARY
Groups Count Sum Average Variance
BMI for C1 51367 1383543 26.93447 24.03897
BMI for C2 9283 264740.8 28.51889 30.70357
BMI for C3 7856 231499.1 29.46781 33.03306

ANOVA
Source of Variation SS df MS F P-value F crit
Between Groups 56236.93 2 28118.47 1082.59 0 2.995863
Within Groups 1779251 68503 25.97333

Total 1835488 68505

Interpretation:
Group Statistics are as follows:
• C1 (Normal): Mean BMI = 26.93
• C2 (Above Normal): Mean BMI = 28.52
• C3 (Well Above Normal): Mean BMI = 29.47

This shows a gradual increase in mean BMI with higher cholesterol levels.

ANOVA Test Results

Since the F-value (1082.59) is much greater than F-critical (2.99) and the p-value < 0.05, we reject the null
hypothesis.

There is a statistically significant difference in BMI across cholesterol categories. In other words,
cholesterol level has a strong influence on BMI. Patients with higher cholesterol levels (C2, C3) tend to
have higher BMI compared to those with normal cholesterol (C1).
Two-Way Anova: While One-Way ANOVA tested if BMI differs across cholesterol levels, it did not consider
the combined effects. Two-Way ANOVA examines whether cholesterol and glucose individually affect
BMI. This is important since both are key metabolic indicators whose joint effect may reveal deeper insights
into body weight and cardiovascular risks.

Variables Used
• Dependent Variable: BMI (continuous).
• Independent Variable (Categorical): Cholesterol level (3 categories: 1 = normal, 2 = above normal,
3 = well above normal) and Glucose level (3 categories: 1 = normal, 2 = above normal, 3 = well
above normal)

ANOVA: Two factor without replication

SUMMARY Count Sum Average Variance
Row 1 (C = 1) 3 83.89582037 27.96527346 1.175849457
Row 2 (C = 2) 3 87.65138252 29.21712751 1.081665493
Row 3 (C = 3) 3 90.20272993 30.06757664 1.841187536

Column 1 (G = 1) 3 84.53175783 28.17725261 1.919034175

Column 2 (G = 2) 3 89.82542018 29.94180673 2.09214377
Column 3 (G = 3) 3 87.39275481 29.13091827 1.102230284

ANOVA
Source of
Variation SS df MS F P-value F crit
Rows 6.710081 2 3.3550405 3.816085159 0.118249 6.944272
Columns 4.68066951 2 2.340334756 2.661940066 0.184046 6.944272
Error 3.51673546 4 0.879183865

Total 14.907486 8

Group Statistics are as follows:

• Row 1 (Cholesterol Level 1): Mean BMI ≈ 27.97, Row 2 (Cholesterol Level 2): Mean BMI ≈ 29.22 and
Row 3 (Cholesterol Level 3): Mean BMI ≈ 30.07
• Column 1 (Glucose Level 1): Mean BMI ≈ 28.18, Column 2 (Glucose Level 2): Mean BMI ≈ 29.94 and
Column 3 (Glucose Level 3): Mean BMI ≈ 29.13
These averages show slight variation across cholesterol and glucose levels, but the statistical test
determines whether these differences are significant.

ANOVA Test Results:

• For Rows (Cholesterol): F = 3.82, F-critical = 6.94, p = 0.118 (> 0.05), suggesting Fail to reject H₀.
• For Columns (Glucose): F = 2.66, F-critical = 6.94, p = 0.184 (> 0.05), suggesting Fail to reject H₀.

There is no statistically significant difference in BMI across cholesterol levels or glucose levels in this test.
In other words, neither cholesterol level nor glucose level alone has a strong influence on BMI when
considered separately. This suggests that other factors (such as lifestyle, genetics, or combined
interactions not captured here) may be stronger determinants of BMI. Please note that, since we used
“without replication,” interaction between cholesterol and glucose could not be tested.

[Link]
27/Term%201/Business%20Statistics/Workings_Anova.xlsx
Chi-Square Test:

Results as per the Test:

Frequency Table

Group High BP Normal BP Total

Weight <100 131 48 179
Weight >100 72 3 75
Total 203 51 254

Chi-Square Table
High BP Normal BP Total
Weight <100 143.06 35.94 179
Weight >100 59.94 15.06 75
Total 203 51 254

Chi-Square 3E-05

Inference
• Very strong association: The extremely low p-value means the distribution of High/Normal BP is
statistically different between Weight < 100 and Weight > 100 groups.
• Chi-square value is extremely small, and the result is statistically significant.
• This means people with more than 100 kg weight are much more likely to have blood pressure
above 120, and this pattern is not just a coincidence.
• Conclusion: Weight is significantly associated with blood pressure status at the 120 BP threshold;
higher weight is linked to a much higher proportion of High BP cases.

Simple Linear Regression

SUMMARY OUTPUT

Regression Statistics
Multiple R 0.27
R Square 0.07
Adjusted R Square 0.07
Standard Error 13.66
Observations 68506.00

ANOVA
Significance
df SS MS F F
Regression 1 1014988.33 1014988.33 5439.53 0
Residual 68504 12782487.64 186.59
Total 68505 13797475.97
 Coefficients Standard Error t Stat P-value Lower 95% Upper 95%
Intercept 44.88 0.40 112.24 0.00 44.10 45.66
ap_hi 0.23 0.00 73.75 0.00 0.22 0.24

This regression output shows how weight predicts blood pressure in your sample of 68,506 people.

Inferences:

• Multiple R (0.27): This shows a weak positive relationship between weight and blood pressure; as
weight increases, blood pressure tends to go up with a little movement.
• R Square (0.07): Only about 7% of the variation in blood pressure can be explained by weight. This
means many other factors are also affecting blood pressure.
• Standard Error (13.66): This means that on average, predictions from the model may be off by about
13.66 units of blood pressure.
Regression Coefficients

• Intercept (44.88): When weight is zero, the model predicts a blood pressure value of 44.88 (which is not
realistic, but just to give a baseline).
• Blood Pressure (0.23): For every increase of 1 unit in weight, predicted blood pressure rises by 0.23
units on average.
• P-values (both 0.00): Both coefficients are statistically significant (highly unlikely to happen by chance).
• Confidence Intervals for Blood Pressure: The true effect is likely between 0.22 and 0.24 units per kg.
To sum it up:

• Heavier people tend to have slightly higher blood pressure, but weight alone doesn't explain blood
pressure well (weak link).
• The model is significant (not due to chance), but other factors play a much bigger role in causing blood
pressure differences.
• In summary: There's a weak, positive, and statistically significant relationship between weight and blood
pressure in the data.

ap_hi
ap_hi Linear (ap_hi) Linear (ap_hi)

300 y = 0.3187x + 103.06

R² = 0.0736
250

200
Weight

150

100

50

0
0 20 40 60 80 100 120 140 160 180

This scatter plot with a trend line shows the relationship between blood pressure (ap_hi) and weight for
hospital patients.

• The orange dots are individual data points, each representing a patient's blood pressure and weight.
• The trend line equation y=0.3187x+103.06y=0.3187x+103.06 shows that as blood pressure (x)
increases, predicted weight (y) also rises slightly.
• The plot confirms a positive but weak relationship—heavier patients generally have higher blood
pressure.
• The R^2 value is 0.0736; meaning that only about 7% of the variation in weight is explained by blood
pressure.
• Most orange points scatter widely around the trend line, illustrating significant overlap and other factors
that influence blood pressure.

Let’s look at how all this data can be used by the hospital from a business point of view:

From a business perspective, analysing hospital data showing a weak but statistically significant
relationship between weight and blood pressure provides several insights and opportunities:

1. Patient Management and Preventive Care

• The data confirms that higher weight is associated with slightly higher blood pressure—a known risk
factor for cardiovascular diseases.
• The hospital can develop targeted weight management and blood pressure monitoring programs to
identify and support patients at risk early, reducing complications and long-term healthcare costs.

2. Service Offerings and Health Packages

• Based on this relationship, the hospital might design specialized health packages focused on obesity-
related risk assessments, including regular blood pressure and weight checks, nutritional counselling,
and fitness programs.
• These packages can be marketed to corporate wellness programs, insurance providers, & individual
patients.

3. Resource Allocation and Operational Planning

• Understanding that weight influences blood pressure helps anticipate the demand for cardiovascular
services, potentially influencing staffing, equipment procurement (e.g., for cardiac care), and budget
planning.
• High-risk patient groups can be prioritized for more intensive monitoring, optimizing resource use.

4. Data-Driven Patient Education

• The hospital can leverage data insights to create educational content emphasizing weight control to
manage blood pressure, encouraging patient adherence to lifestyle interventions.
• Increased patient engagement can improve long-term outcomes and reduce readmissions.

5. Research and Collaboration

• The hospital can use this evidence to collaborate with insurers, pharmaceutical companies, or public
health agencies to design risk-reduction trials or community health programs, potentially unlocking new
funding or partnership opportunities.
6. Revenue Growth through Personalized Care
• Personalizing care based on weight and blood pressure correlations supports precision medicine,
potentially attracting more patients seeking evidence-based management, leading to revenue growth
and enhanced reputation.
To sum it all up despite the relationship being weak, the hospital can capitalize on this knowledge to
enhance preventive care, optimize operations, design targeted services, and engage patients effectively, all
contributing to better health outcomes and business growth.
Multiple Regression Model

Model 1: One Dependent and three explanatory variables

Dependent Variable (Y): ap_hi (Systolic BP)

Independent Variables (X): X1: Age, X2: BMI, X3: ap_lo

ap_hi = β0 + β1⋅Age + β2⋅BMI + β3⋅ap_lo + ϵ

or

Y= β0 + β1⋅X1 + β2⋅X2 + β3⋅X3 + ϵ

Y^ (Y hat) = b0 + b1 X1+ b2 X2+ b3 X3

Where b0, b1, b2, and b3 are estimators of β0, β1, β2 and β3 respectively.

A-priori expectations (signs and interpretation)

• β₁ (Age → ap_hi): Expected sign positive as systolic BP typically rises with age (arterial stiffening etc.)
• β₂ (BMI → ap_hi): Expected sign positive since higher BMI is associated with higher systolic BP.
• β₃ (ap_lo → ap_hi): Expected sign positive (strongly) since systolic and diastolic pressures are
correlated physiologically; therefore higher diastolic tends to fo together with higher systolic.

Running the Regression by Excel

Regression Statistics
Multiple R 0.745
R Square 0.555
Adjusted R Square 0.555
Standard Error 11.129
Observations 68506

ANOVA
df SS MS F Significance F
Regression 3 10569417.451 3523139.150 28447.006 0
Residual 68502 8483918.547 123.849
Total 68505 19053336

Lower Upper
Coefficients Standard Error t Stat P-value 95% 95%
Intercept 6.478 0.480 13.509 1.562E-41 5.538 7.418
age (years) 0.225 0.006 35.203 4.667E-269 0.212 0.237
BMI 0.293 0.008 34.554 2.124E-259 0.277 0.310
ap_lo 1.232 0.005 262.670 0.000E+00 1.223 1.241

Regression Line ap_hi^ = 6.478 + 0.225⋅Age + 0.293⋅BMI + 1.232⋅ap_lo

Interpretation of Regression Results

• The model explains approximately 55.5% of the variation in systolic blood pressure (R² = 0.555).
• F-test for overall significance is highly significant (p < 0.001), indicating the model provides a good fit.
1. Intercept (b = 6.478): The model predicts that when Age, BMI & ap_lo are all zero, the expected
systolic BP (ap_hi) would be 6.48 mmHg. However, such situation is not realistic in practice, so the
intercept mainly serves as baseline constant that adjusts the regression line for accurate prediction.

2. Age (b = 0.225): For each additional year of age, ap_hi increases by 0.23 mmHg, holding BMI & ap_lo
constant. This aligns with medical evidence that blood pressure tends to rise with age.

3. BMI (b = 0.293): For every one-unit increase in BMI, systolic blood pressure (ap_hi) increases by about
0.29 mmHg, keeping other variables like age and ap_lo constant. This suggests that higher body mass
index (overweight/obesity) is positively associated with elevated systolic blood pressure.

4. Diastolic blood pressure, ap_lo (b = 1.232): For every 1 mmHg increase in diastolic blood pressure
(ap_lo), systolic blood pressure increases by about 1.23 mmHg, keeping Age and BMI constant. This
strong association is expected since systolic and diastolic blood pressures are physiologically related.

Conclusion: All three predictors are statistically significant (p<0.05) and positively associated with systolic
blood pressure. Among them, diastolic blood pressure (ap_lo) has the largest effect, while Age and BMI
also contribute meaningfully. The results highlight that older age, higher BMI, and higher diastolic BP are
important risk factors linked to higher systolic BP.

Overall Model Test (F-test)

Goal: To test whether the regression model with the three explanatory variables explains a significant
portion of the variance in systolic blood pressure.

Hypotheses:
H0 : β1 = β2 = β3 = 0
HA : At least one βj ≠ 0

Test statistic: F = MSRegression/ MSResidual where MS= mean square.

Excel’s regression output provides F and Significance F.

Results: Overall F(3, 68502) = 28447.01, p < 0.05 → reject H0

Conclusion: The regression model with Age, BMI, and ap_lo explains a statistically significant portion of
the variation in systolic blood pressure. In other words, taken together the three predictors provide
explanatory power beyond the intercept alone (R² = 0.555), and the overall model is statistically
significant at the 5% level

Individual coefficient tests (t-tests)

Goal: To test if each coefficient is significant or not (Age, BMI, ap_lo).

Hypotheses (two-tailed):
H0 : βj = 0
HA : βj ≠ 0

Test statistic: t = bj / SE(bj)

Decision Rule: Reject H0 if p-value < 0.05

Age: t(68502) = 35.20, p<0.05 → reject H0

BMI: t(68502) = 34.55, p<0.05 → reject H0
ap_lo: t(68502) = 262.67, p<0.05 → reject H0

Conclusion: Age, BMI, and diastolic blood pressure (ap_lo) are all statistically significant predictors of
systolic blood pressure at the 5% level.
Contributions Topics Covered
Focused on multiple sample hypothesis testing such as one-way/two-way ANOVA.
Vanshika

Worked on one-sample hypothesis testing including z-tests, t-tests, one-tailed tests,

Mohit and proportion tests.
Contributed by covering sampling methods, sampling distributions, central limit theorem,
Pratyush confidence intervals, and sample size determination.

Took up regression and correlation topics including covariance, correlation matrix, and
Tisha multiple linear regression with categorical variables.
Handled two-sample hypothesis testing comparing means, proportions, and variance
Yuvraj ratios (F-test).
Covered non-parametric tests including Chi-square tests for difference in proportions and
Kartikey independence, Simple Linear Regression.