Code
library(tidyverse)
library(readxl)
library(janitor)
library(corrplot)
library(ggcorrplot)
library(car)
library(pROC)
library(ResourceSelection)
library(knitr)
library(kableExtra)
library(scales)
library(readr)
library(patchwork)Understanding the Lagos Children’s Event Market: A Data-Driven Analysis for Sweet Indulgence by Hobams
Code
# Load raw data
df_raw <- read_excel("sih_data_to_be_used.xlsx") %>%
clean_names()
# Rename columns to short readable names
df <- df_raw %>%
rename(
timestamp = timestamp,
num_children = how_many_children_do_you_have,
age_groups = what_are_the_age_groups_of_your_children_tick_all_that_apply,
area = which_area_of_lagos_are_you_based_in,
income = what_is_your_monthly_household_income_range,
events_per_year = how_many_childrens_events_do_you_host_per_year,
event_types = what_types_of_childrens_events_do_you_host_tick_all_that_apply,
budget_raw = what_is_your_typical_total_budget_per_childrens_event,
treat_types = which_treat_types_do_your_children_enjoy_most_at_events_select_up_to_3,
treat_pct = what_percentage_of_your_event_budget_goes_to_treats_or_food_for_children,
vendor_used = have_you_used_a_professional_treat_vendor_at_a_childrens_event_before,
satisfaction = how_satisfied_were_you_with_that_vendor_rate_from_1_very_dissatisfied_to_10_extremely_satisfied,
find_vendors = how_do_you_typically_find_childrens_event_vendors_tick_all_that_apply,
premium_willing = would_you_pay_a_premium_for_a_dedicated_childrens_treat_vendor_with_custom_branding_and_themed_setups,
frustration = what_is_your_biggest_frustration_or_challenge_when_organizing_treats_for_childrens_events,
unforgettable = what_would_make_a_childrens_treat_experience_truly_unforgettable_for_your_family
) %>%
# Drop Score column - 100% missing, never activated in Google Forms
select(-score) %>%
# Fix events_per_year - Google Forms stored number 2 as a date
mutate(events_per_year = case_when(
grepl("2026", as.character(events_per_year)) ~ "2",
TRUE ~ as.character(events_per_year)
)) %>%
# Clean budget column - free text with 61 different formats
mutate(budget_cleaned = parse_number(
budget_raw, locale = locale(grouping_mark = ",")
)) %>%
mutate(budget_cleaned = case_when(
grepl("1\\.5m|1\\.5M|1,500,000", budget_raw, ignore.case = TRUE) ~ 1500000,
grepl("don|dnt|depends|No answer|Less than",
budget_raw, ignore.case = TRUE) ~ NA_real_,
TRUE ~ budget_cleaned
)) %>%
# Encode ordinal variables as numbers for correlation and regression
mutate(
income_encoded = case_when(
grepl("Below", income, ignore.case = TRUE) ~ 1,
grepl("200,000 - 500|200k - 500", income, ignore.case = TRUE) ~ 2,
grepl("500,000 - 1|500k - 1", income, ignore.case = TRUE) ~ 3,
grepl("Above|1,000,000", income, ignore.case = TRUE) ~ 4,
TRUE ~ NA_real_
),
treat_pct_encoded = case_when(
grepl("less than 10|<10", treat_pct, ignore.case = TRUE) ~ 1,
grepl("10.*20|10-20", treat_pct, ignore.case = TRUE) ~ 2,
grepl("20.*30|20-30", treat_pct, ignore.case = TRUE) ~ 3,
grepl("more than 30|>30", treat_pct, ignore.case = TRUE) ~ 4,
TRUE ~ NA_real_
),
premium_encoded = case_when(
grepl("No", premium_willing, ignore.case = TRUE) ~ 0,
grepl("Maybe", premium_willing, ignore.case = TRUE) ~ 1,
grepl("Yes", premium_willing, ignore.case = TRUE) ~ 2,
TRUE ~ NA_real_
),
premium_yes = if_else(
grepl("Yes", premium_willing, ignore.case = TRUE), 1, 0
),
vendor_used_num = if_else(
grepl("Yes", vendor_used, ignore.case = TRUE), 1, 0
),
area_encoded = if_else(
grepl("Island|Lekki|Victoria", area, ignore.case = TRUE), 1, 0
),
num_children_encoded = case_when(
grepl("^1$", num_children) ~ 1,
grepl("^2$", num_children) ~ 2,
grepl("^3$", num_children) ~ 3,
grepl("4|more", num_children, ignore.case = TRUE) ~ 4,
TRUE ~ NA_real_
),
satisfaction = as.numeric(satisfaction)
)
# Confirm it worked
cat("Rows:", nrow(df), "\nColumns:", ncol(df))Rows: 100
Columns: 24# 1. Executive Summary
This study applies five exploratory and inferential analytical techniques to a primary dataset of 100 survey responses collected from Lagos-based parents by Sweet Indulgence by Hobams, Lagos’s premier kiddies treat experience brand. The dataset was collected via Google Forms in May 2026 across 17 variables covering demographics, event behaviour, treat preferences, vendor experience, and premium willingness.
Exploratory data analysis revealed that 51% of respondents allocate more than 30% of their event budget to treats, confirming treats as a primary spend category. Ice cream and popcorn are the most popular treat types, validating the brand’s core product offering. Data visualisation identified a progressive relationship between income and premium willingness. Hypothesis testing found no statistically significant difference in vendor satisfaction between Lagos Island and Mainland parents (p = 0.737), and no significant association between income and premium willingness (p = 0.678). Correlation analysis identified treat budget percentage as the strongest correlate of premium willingness. Logistic regression confirmed that treat-conscious parents — those allocating a higher percentage of their budget to treats — are the strongest predictor of premium uptake, ahead of income or location.
The unified recommendation is that Sweet Indulgence by Hobams should prioritise treat-conscious parents in its marketing strategy, targeting those who already allocate 20% or more of their event budget to treats, regardless of income bracket or Lagos location. This customer segment offers the highest probability of premium conversion and is reachable through the brand’s existing Instagram and WhatsApp marketing channels.
# 2. Professional Disclosure
My name is Adeyinka Adedoyin Obabiolorunkosi. I am the Founder and Chief Experience Officer of Sweet Indulgence by Hobams — Lagos’s premier kiddies treat experience brand — and concurrently serve as Territorial Sales Manager at Multichoice Nigeria, one of Africa’s largest entertainment companies. My professional journey is anything but conventional: it began in the enclosures of Origin Gardens Beach & Zoo in Lagos and the National Children Zoo & Park in Abuja, where I guided children through worlds of wonder and first understood the profound power of purposeful, joyful experiences for young people. That early revelation has shaped every career decision since.
Over fifteen years across zoo operations, animal welfare NGOs, insurance sales, agricultural development, creative direction, and corporate sales leadership, I have built one consistent skill: understanding what people want and designing experiences that deliver it. Today, that skill is the engine of Sweet Indulgence by Hobams — a brand that brings premium popcorn, ice cream, popsicles and cotton candy to children’s parties, school events, corporate family days and community activations across Lagos.
It is within this entrepreneurial context that this data analytics study was conceived. As a founder making real decisions about pricing, product mix, market positioning, and customer targeting with limited resources, I cannot afford to rely on intuition alone. I need evidence. The five analytical techniques applied in this study are not academic exercises selected to satisfy an assignment brief — they are tools I genuinely need to grow this business:
Exploratory Data Analysis answers my most basic but critical question: who are my customers and what do they actually want? As a founder who distributes through WhatsApp and Instagram, I need to understand the demographics, preferences, and behaviours of the Lagos parents I serve before I can scale. EDA gives me that foundation.
Data Visualisation is how I communicate market insights to the people who matter — potential investors, corporate sponsors, the Hobams Foundation board, and the National Children Zoo & Park Abuja, with whom I hold a formal strategic partnership. A clean, compelling chart does what a paragraph of text cannot: it makes a decision obvious.
Hypothesis Testing addresses a question that affects my entire go-to-market strategy: do Lagos Island and Mainland parents behave differently? If they do, I need differentiated pricing and service tiers. If they do not, I can standardise my offering and scale more efficiently. This is not a theoretical question — it is one I face every time I price an activation package.
Correlation Analysis helps me identify which customer characteristics move together, so I can build a precise profile of my ideal premium customer rather than marketing to everyone and converting few. In a bootstrapped startup with no dedicated marketing budget, targeting precision is not a luxury — it is survival.
Logistic Regression is the most powerful tool in this study for my business. It tells me, with statistical rigour, which combination of customer characteristics predicts premium willingness. That prediction model is, in practical terms, a customer scoring system — one that tells my sales effort exactly where to focus first.
Every dataset, every hypothesis, every result in this document maps to a decision I am actively navigating as a founder. That is the spirit in which this study was conducted.
# 3. Data Collection & Sampling
## Source & Collection Method
The dataset was collected via a structured Google Forms survey designed and
administered by the researcher in her capacity as Founder of Sweet Indulgence
by Hobams. The survey was distributed through WhatsApp groups, Instagram direct
messages and Instagram Stories — the same channels used to market the brand —
between June and December 2025.
## Sampling Frame
The target population is Lagos-based parents with at least one child who
organise or attend children’s events. This population is the direct target
customer for Sweet Indulgence by Hobams. A purposive convenience sampling
method was used, appropriate because the brand’s market is socially-connected
Lagos parents rather than a random general population.
## Sample Size & Statistical Rationale
A total of 100 complete responses were collected across 17 variables. By the
Central Limit Theorem, n ≥ 30 is sufficient for sampling distributions of means
to be approximately normal; n = 100 provides adequate statistical power for the
t-tests, chi-squared tests, correlation analyses and logistic regression planned
in this study.
## Ethical Notes
No personally identifiable information was collected. Participation was
voluntary. Respondents were informed the data would be used for academic and
business research purposes. No formal ethics board approval was required as the
survey collected no sensitive personal data.
## Data Quality Issues Identified
Three data quality issues were identified and resolved during cleaning:
1. **Budget variable**: collected as free text, producing 61 unique
non-standardised formats (e.g. “500,000”, “500k”, “1.5m”). Resolved using
`parse_number()` with manual override for edge cases.
2. **Events per year**: Google Forms misread numeric input “2” as a date
(“2026-02-03”). Resolved by detecting and recoding date-formatted entries.
3. **Score column**: contained zero non-null values across all 100 rows — the
Google Forms quiz-scoring feature was never activated. Column excluded from
all analyses.
# 4. Data Description
*EDA code and output go here — next step*
# 5. Technique 1 — Exploratory Data Analysis
Exploratory Data Analysis (EDA) is the foundation of any rigorous analytical workflow. Before fitting models or running tests, a practitioner must understand the shape, distribution, and quality of their data. For Sweet Indulgence by Hobams, EDA answers the most fundamental business question: who are my customers and what do they look like?
Code
# Summary statistics for all numeric variables
df %>%
select(num_children_encoded, income_encoded, treat_pct_encoded,
premium_encoded, vendor_used_num, area_encoded,
satisfaction, budget_cleaned) %>%
summary() num_children_encoded income_encoded treat_pct_encoded premium_encoded
Min. :1.0 Min. :1.000 Min. :1.00 Min. :0.00
1st Qu.:2.0 1st Qu.:1.750 1st Qu.:2.75 1st Qu.:1.00
Median :2.0 Median :4.000 Median :4.00 Median :1.00
Mean :2.3 Mean :3.222 Mean :3.22 Mean :1.15
3rd Qu.:3.0 3rd Qu.:4.000 3rd Qu.:4.00 3rd Qu.:2.00
Max. :4.0 Max. :4.000 Max. :4.00 Max. :2.00
NA's :46
vendor_used_num area_encoded satisfaction budget_cleaned
Min. :0.00 Min. :0.00 Min. : 1.000 Min. : 1
1st Qu.:0.00 1st Qu.:0.00 1st Qu.: 5.000 1st Qu.: 30000
Median :1.00 Median :1.00 Median : 7.000 Median : 150000
Mean :0.66 Mean :0.56 Mean : 6.687 Mean : 367117
3rd Qu.:1.00 3rd Qu.:1.00 3rd Qu.: 8.000 3rd Qu.: 500000
Max. :1.00 Max. :1.00 Max. :10.000 Max. :3000000
NA's :17 NA's :11 Code
# Missing value analysis
missing_summary <- df %>%
summarise(across(everything(), ~ sum(is.na(.)))) %>%
pivot_longer(everything(),
names_to = "Variable",
values_to = "Missing_Count") %>%
mutate(Missing_Pct = round(Missing_Count / nrow(df) * 100, 1)) %>%
filter(Missing_Count > 0) %>%
arrange(desc(Missing_Count))
missing_summary %>%
kable(caption = "Table 1: Missing Values by Variable") %>%
kable_styling(bootstrap_options = c("striped", "hover"))| Variable | Missing_Count | Missing_Pct |
|---|---|---|
| income_encoded | 46 | 46 |
| satisfaction | 17 | 17 |
| events_per_year | 14 | 14 |
| budget_cleaned | 11 | 11 |
| frustration | 2 | 2 |
| budget_raw | 1 | 1 |
Code
# Distribution of key categorical variables
p1 <- df %>%
count(income) %>%
mutate(income = str_wrap(income, 20)) %>%
ggplot(aes(x = reorder(income, n), y = n, fill = income)) +
geom_col(show.legend = FALSE) +
coord_flip() +
labs(title = "Income Distribution of Respondents",
x = "Income Range", y = "Count") +
theme_minimal()
p2 <- df %>%
count(area) %>%
mutate(area = str_wrap(area, 20)) %>%
ggplot(aes(x = reorder(area, n), y = n, fill = area)) +
geom_col(show.legend = FALSE) +
coord_flip() +
labs(title = "Respondents by Lagos Area",
x = "Area", y = "Count") +
theme_minimal()
p3 <- df %>%
count(events_per_year) %>%
ggplot(aes(x = reorder(events_per_year, n), y = n, fill = events_per_year)) +
geom_col(show.legend = FALSE) +
coord_flip() +
labs(title = "Events Hosted Per Year",
x = "Events Per Year", y = "Count") +
theme_minimal()
p4 <- df %>%
filter(!is.na(budget_cleaned)) %>%
ggplot(aes(x = budget_cleaned)) +
geom_histogram(bins = 20, fill = "#2C7BB6", colour = "white") +
scale_x_continuous(labels = comma) +
labs(title = "Distribution of Event Budget (₦)",
x = "Budget (₦)", y = "Count") +
theme_minimal()
library(patchwork)
(p1 + p2) / (p3 + p4)
Code
# Outlier detection using boxplot on budget and satisfaction
p5 <- df %>%
filter(!is.na(budget_cleaned)) %>%
ggplot(aes(y = budget_cleaned)) +
geom_boxplot(fill = "#FDB863") +
scale_y_continuous(labels = comma) +
labs(title = "Boxplot: Event Budget (₦)",
y = "Budget (₦)") +
theme_minimal()
p6 <- df %>%
filter(!is.na(satisfaction)) %>%
ggplot(aes(y = satisfaction)) +
geom_boxplot(fill = "#74ADD1") +
labs(title = "Boxplot: Vendor Satisfaction Score",
y = "Satisfaction (1-10)") +
theme_minimal()
p5 + p6
Key EDA Findings:
- The dataset contains 100 observations across 16 active variables after removing the empty Score column.
- Satisfaction scores are missing for 17 respondents — these are parents who have never used a professional vendor, so the missingness is structural and expected, not random.
- Budget is missing for a small number of respondents who entered non-numeric responses such as “Don’t have one” — these were coded as NA.
- The budget boxplot reveals at least one high-value outlier (₦1,500,000+), indicating a small segment of very high-spending customers — important for Sweet Indulgence’s premium tier planning.
- Income is skewed toward the middle brackets, with the largest group falling in the ₦200,000–₦500,000 range.
# 6. Technique 2 — Data Visualisation
Effective data visualisation transforms numbers into decisions. For Sweet Indulgence by Hobams, the goal of this visualisation narrative is to answer one central business question: who is my customer and what drives their willingness to pay a premium? Five complementary plots are presented below, each chosen deliberately for its ability to communicate a specific pattern in the data.
Code
# Plot 1 - Treat type popularity (explode multi-select column)
df %>%
filter(!is.na(treat_types)) %>%
separate_rows(treat_types, sep = ",") %>%
mutate(treat_types = str_trim(treat_types)) %>%
count(treat_types, sort = TRUE) %>%
ggplot(aes(x = reorder(treat_types, n), y = n, fill = treat_types)) +
geom_col(show.legend = FALSE) +
coord_flip() +
labs(
title = "Figure 1: Most Popular Treat Types at Children's Events",
subtitle = "Multiple selections allowed — n = 100 respondents",
x = "Treat Type",
y = "Number of Mentions",
caption = "Source: Sweet Indulgence by Hobams Customer Survey, 2025"
) +
theme_minimal(base_size = 12)
Business interpretation (Figure 1): Ice cream and popcorn dominate treat preferences among Lagos parents. For Sweet Indulgence by Hobams, this confirms that the core product offering — premium popcorn and ice cream — is correctly aligned with market demand. Cotton candy and popsicles serve as complementary upsell products.
Code
# Plot 2 - Income group vs premium willingness (stacked bar)
df %>%
filter(!is.na(income), !is.na(premium_willing)) %>%
mutate(income = str_wrap(income, 25)) %>%
count(income, premium_willing) %>%
group_by(income) %>%
mutate(pct = n / sum(n) * 100) %>%
ggplot(aes(x = income, y = pct, fill = premium_willing)) +
geom_col(position = "stack") +
coord_flip() +
scale_fill_brewer(palette = "RdYlGn", direction = 1) +
labs(
title = "Figure 2: Premium Willingness by Income Group",
subtitle = "Percentage within each income bracket",
x = "Income Range",
y = "Percentage (%)",
fill = "Premium Willingness",
caption = "Source: Sweet Indulgence by Hobams Customer Survey, 2025"
) +
theme_minimal(base_size = 12) +
theme(legend.position = "bottom")
Business interpretation (Figure 2): Higher income brackets show a progressively larger share of “Yes, definitely” responses. This gives Sweet Indulgence by Hobams a clear targeting signal — premium branded packages should be marketed primarily to households earning above ₦500,000 per month.
Code
# Plot 3 - Vendor satisfaction by area (boxplot)
df %>%
filter(!is.na(satisfaction), !is.na(area)) %>%
mutate(area = str_wrap(area, 20)) %>%
ggplot(aes(x = reorder(area, satisfaction, median),
y = satisfaction, fill = area)) +
geom_boxplot(show.legend = FALSE, alpha = 0.7) +
coord_flip() +
labs(
title = "Figure 3: Vendor Satisfaction Scores by Lagos Area",
subtitle = "Among the 83 respondents who have used a vendor before",
x = "Lagos Area",
y = "Satisfaction Score (1–10)",
caption = "Source: Sweet Indulgence by Hobams Customer Survey, 2025"
) +
theme_minimal(base_size = 12)
Business interpretation (Figure 3): Satisfaction scores vary across Lagos areas, with some areas showing wider spread and lower medians — indicating underserved markets where current vendor quality is inconsistent. These areas represent the strongest opportunity for Sweet Indulgence by Hobams to enter and capture dissatisfied customers.
Code
# Plot 4 - Treat budget percentage (bar chart)
df %>%
filter(!is.na(treat_pct)) %>%
count(treat_pct) %>%
mutate(treat_pct = factor(treat_pct,
levels = c("Less than 10%", "10 - 20%", "20 - 30%", "More than 30%"))) %>%
ggplot(aes(x = treat_pct, y = n, fill = treat_pct)) +
geom_col(show.legend = FALSE) +
geom_text(aes(label = n), vjust = -0.5, size = 4) +
labs(
title = "Figure 4: Treat Spend as Percentage of Event Budget",
subtitle = "51 of 100 respondents allocate more than 30% of their budget to treats",
x = "Treat Budget Percentage",
y = "Number of Respondents",
caption = "Source: Sweet Indulgence by Hobams Customer Survey, 2025"
) +
theme_minimal(base_size = 12)
Business interpretation (Figure 4): The majority of respondents — 51 out of 100 — allocate more than 30% of their event budget to treats. This demolishes the assumption that treats are an afterthought at children’s events. For Sweet Indulgence by Hobams, treats are a primary spend category, not a secondary one. This justifies premium pricing.
Code
# Plot 5 - Budget vs satisfaction scatter
df %>%
filter(!is.na(budget_cleaned), !is.na(satisfaction)) %>%
ggplot(aes(x = budget_cleaned, y = satisfaction,
colour = vendor_used)) +
geom_point(alpha = 0.7, size = 3) +
geom_smooth(method = "lm", se = TRUE, colour = "black",
linetype = "dashed") +
scale_x_continuous(labels = comma) +
scale_colour_manual(values = c("Yes" = "#2C7BB6", "No" = "#D7191C")) +
labs(
title = "Figure 5: Event Budget vs Vendor Satisfaction Score",
subtitle = "Coloured by whether a professional vendor was used",
x = "Total Event Budget (₦)",
y = "Vendor Satisfaction Score (1–10)",
colour = "Used Professional Vendor",
caption = "Source: Sweet Indulgence by Hobams Customer Survey, 2025"
) +
theme_minimal(base_size = 12) +
theme(legend.position = "bottom")
Business interpretation (Figure 5): There is a visible positive trend between event budget and vendor satisfaction — higher-spending parents tend to report higher satisfaction with their vendors. This scatter plot sets up the formal correlation analysis in Section 8 and supports the regression model in Section 9. The dashed trend line confirms the direction of the relationship before formal testing.
# 7. Technique 3 — Hypothesis Testing
Hypothesis testing allows a business analyst to move beyond description and make statistically defensible decisions. For Sweet Indulgence by Hobams, two business questions are tested formally below. Each test follows the full protocol: state hypotheses, check assumptions, run the test, report the p-value and effect size, and interpret the result in plain business language.
Test 1 — Welch t-test: Does vendor satisfaction differ between Lagos Island and Mainland?
Business justification: If Island and Mainland parents experience significantly different satisfaction levels with existing vendors, Sweet Indulgence by Hobams should develop differentiated service offerings for each market rather than a single one-size-fits-all approach.
H₀: Mean vendor satisfaction is the same for Island and Mainland parents (μ₁ = μ₂)
H₁: Mean vendor satisfaction differs between Island and Mainland parents (μ₁ ≠ μ₂)
Code
# Prepare data - filter to vendor users only and split by area
island <- df %>%
filter(!is.na(satisfaction),
grepl("Island|Lekki|Victoria", area, ignore.case = TRUE)) %>%
mutate(satisfaction = as.numeric(satisfaction)) %>%
pull(satisfaction)
mainland <- df %>%
filter(!is.na(satisfaction),
!grepl("Island|Lekki|Victoria", area, ignore.case = TRUE)) %>%
pull(satisfaction)
cat("Island n =", length(island),
"| Mean =", round(mean(island), 2), "\n")Island n = 42 | Mean = 6.76 Code
cat("Mainland n =", length(mainland),
"| Mean =", round(mean(mainland), 2), "\n\n")Mainland n = 41 | Mean = 6.61 Code
# Assumption check 1 - Shapiro-Wilk normality test
cat("=== Shapiro-Wilk Normality Test ===\n")=== Shapiro-Wilk Normality Test ===Code
shapiro.test(island)
Shapiro-Wilk normality test
data: island
W = 0.93128, p-value = 0.01431Code
shapiro.test(mainland)
Shapiro-Wilk normality test
data: mainland
W = 0.93448, p-value = 0.0204Code
# Assumption check 2 - Levene's test for equal variances
cat("\n=== Levene's Test for Equal Variances ===\n")
=== Levene's Test for Equal Variances ===Code
sat_data <- df %>%
filter(!is.na(satisfaction)) %>%
mutate(area_group = if_else(
grepl("Island|Lekki|Victoria", area, ignore.case = TRUE),
"Island", "Mainland"))
leveneTest(satisfaction ~ area_group, data = sat_data)Levene's Test for Homogeneity of Variance (center = median)
Df F value Pr(>F)
group 1 0.9972 0.321
81 Code
# Run Welch t-test (does not assume equal variances)
t_result <- t.test(island, mainland, var.equal = FALSE)
print(t_result)
Welch Two Sample t-test
data: island and mainland
t = 0.33673, df = 76.948, p-value = 0.7372
alternative hypothesis: true difference in means is not equal to 0
95 percent confidence interval:
-0.747590 1.051887
sample estimates:
mean of x mean of y
6.761905 6.609756 Code
# Effect size - Cohen's d
n1 <- length(island); n2 <- length(mainland)
s_pooled <- sqrt(((n1-1)*var(island) + (n2-1)*var(mainland)) / (n1+n2-2))
cohens_d <- (mean(island) - mean(mainland)) / s_pooled
cat("\nCohen's d =", round(cohens_d, 3))
Cohen's d = 0.074Code
cat("\nEffect size interpretation:",
ifelse(abs(cohens_d) < 0.2, "Negligible",
ifelse(abs(cohens_d) < 0.5, "Small",
ifelse(abs(cohens_d) < 0.8, "Medium", "Large"))))
Effect size interpretation: NegligibleCode
# Visualise the comparison
sat_data %>%
ggplot(aes(x = area_group, y = satisfaction, fill = area_group)) +
geom_boxplot(alpha = 0.7, show.legend = FALSE) +
geom_jitter(width = 0.1, alpha = 0.4, size = 2) +
labs(
title = "Figure 6: Vendor Satisfaction by Lagos Area",
subtitle = "Welch t-test result shown below",
x = "Lagos Area",
y = "Satisfaction Score (1–10)",
caption = "Source: Sweet Indulgence by Hobams Customer Survey, 2025"
) +
theme_minimal(base_size = 12)
Plain-language interpretation: The Welch t-test produced t(76.95) = 0.337, p = 0.737. Since p = 0.737 is well above the 0.05 significance threshold, we fail to reject H₀. There is no statistically significant difference in vendor satisfaction between Lagos Island and Mainland parents. The effect size (Cohen’s d = 0.074) is negligible, confirming the difference is not practically meaningful either. In practical terms for Sweet Indulgence by Hobams, this means a single service standard is appropriate across both markets — the brand does not need to develop separate service offerings by location. However, Island parents report a slightly higher mean satisfaction (6.76 vs 6.61), which may warrant monitoring as the brand scales.
Test 2 — Chi-squared: Is premium willingness associated with income group?
Business justification: If premium willingness is independent of income, Sweet Indulgence by Hobams must rely on factors other than income to identify premium customers. If they are associated, income becomes a valid targeting variable.
H₀: Premium willingness is independent of income range
H₁: There is a significant association between income range and premium willingness
Code
# Build contingency table
cont_table <- table(df$income, df$premium_willing)
print(cont_table)
Maybe, depends on price
#200,000 - #500,000 15
#500,000 - #1,000,000 6
Above #1,000,000 12
Below #200,000 8
Prefer not to say 6
No,i would rather manage it myself Yes,definitely
#200,000 - #500,000 5 9
#500,000 - #1,000,000 3 4
Above #1,000,000 3 12
Below #200,000 4 2
Prefer not to say 4 7Code
# Check expected frequencies - all should be >= 5
cat("\n=== Expected Cell Frequencies ===\n")
=== Expected Cell Frequencies ===Code
chisq_check <- chisq.test(cont_table)
print(round(chisq_check$expected, 1))
Maybe, depends on price
#200,000 - #500,000 13.6
#500,000 - #1,000,000 6.1
Above #1,000,000 12.7
Below #200,000 6.6
Prefer not to say 8.0
No,i would rather manage it myself Yes,definitely
#200,000 - #500,000 5.5 9.9
#500,000 - #1,000,000 2.5 4.4
Above #1,000,000 5.1 9.2
Below #200,000 2.7 4.8
Prefer not to say 3.2 5.8Code
# Run chi-squared test
chi_result <- chisq.test(cont_table)
print(chi_result)
Pearson's Chi-squared test
data: cont_table
X-squared = 5.7222, df = 8, p-value = 0.6783Code
# Effect size - Cramer's V
n <- sum(cont_table)
k <- min(nrow(cont_table), ncol(cont_table))
cramers_v <- sqrt(chi_result$statistic / (n * (k - 1)))
cat("\nCramér's V =", round(cramers_v, 3))
Cramér's V = 0.169Code
cat("\nEffect size interpretation:",
ifelse(cramers_v < 0.1, "Negligible",
ifelse(cramers_v < 0.3, "Small",
ifelse(cramers_v < 0.5, "Medium", "Large"))))
Effect size interpretation: SmallCode
# Visualise association
df %>%
filter(!is.na(income), !is.na(premium_willing)) %>%
mutate(income = str_wrap(income, 25)) %>%
count(income, premium_willing) %>%
group_by(income) %>%
mutate(pct = n / sum(n) * 100) %>%
ggplot(aes(x = income, y = pct, fill = premium_willing)) +
geom_col(position = "dodge") +
coord_flip() +
scale_fill_brewer(palette = "Set2") +
labs(
title = "Figure 7: Premium Willingness by Income Group",
subtitle = "Percentage within each income bracket",
x = "Income Range",
y = "Percentage (%)",
fill = "Premium Willingness",
caption = "Source: Sweet Indulgence by Hobams Customer Survey, 2025"
) +
theme_minimal(base_size = 12) +
theme(legend.position = "bottom")
Plain-language interpretation: The chi-squared test produced χ²(8) = 5.722, p = 0.678. Since p = 0.678 is well above the 0.05 threshold, we fail to reject H₀. Premium willingness is not significantly associated with income range in this sample. Cramér’s V = 0.169 indicates a small and statistically insignificant association. Note: the warning “Chi-squared approximation may be incorrect” indicates some expected cell frequencies fell below 5 — this is acknowledged as a limitation of the small sample within each income-premium cell. In practical terms for Sweet Indulgence by Hobams, income alone is not a reliable predictor of premium willingness. This is an important finding — it means the brand cannot simply target high-income households and expect premium uptake. Other factors such as prior vendor experience, event type, and treat budget allocation must also be considered. This finding directly motivates the logistic regression in Section 9, where multiple predictors are tested simultaneously.
# 8. Technique 4 — Correlation Analysis
Correlation analysis measures the strength and direction of relationships between variables. For Sweet Indulgence by Hobams, this technique answers the question: which customer characteristics move together, and what does that tell us about our ideal premium customer? Spearman correlation is used here rather than Pearson because most variables are ordinal — Spearman makes no assumption of normality and is appropriate for ranked data.
Code
# Select numeric variables for correlation
cor_df <- df %>%
select(
num_children_encoded,
income_encoded,
treat_pct_encoded,
premium_encoded,
vendor_used_num,
area_encoded,
satisfaction,
budget_cleaned
) %>%
rename(
`Num Children` = num_children_encoded,
`Income` = income_encoded,
`Treat Budget %` = treat_pct_encoded,
`Premium Willing` = premium_encoded,
`Used Vendor` = vendor_used_num,
`Island Area` = area_encoded,
`Satisfaction` = satisfaction,
`Event Budget` = budget_cleaned
)
# Compute Spearman correlation matrix
cor_matrix <- cor(cor_df, method = "spearman", use = "pairwise.complete.obs")
# Plot heatmap
ggcorrplot(cor_matrix,
method = "circle",
type = "lower",
lab = TRUE,
lab_size = 3,
colors = c("#D7191C", "white", "#2C7BB6"),
title = "Figure 8: Spearman Correlation Matrix",
ggtheme = theme_minimal())
Code
# Show top correlations in a table
cor_long <- as.data.frame(as.table(cor_matrix)) %>%
rename(Variable1 = Var1, Variable2 = Var2, Correlation = Freq) %>%
dplyr::filter(Variable1 != Variable2) %>%
mutate(Abs_Cor = abs(Correlation)) %>%
arrange(desc(Abs_Cor)) %>%
distinct(Abs_Cor, .keep_all = TRUE) %>%
head(6) %>%
select(Variable1, Variable2, Correlation) %>%
mutate(Correlation = round(Correlation, 3))
cor_long %>%
kable(caption = "Table 2: Top 6 Strongest Spearman Correlations") %>%
kable_styling(bootstrap_options = c("striped", "hover"))| Variable1 | Variable2 | Correlation |
|---|---|---|
| Satisfaction | Used Vendor | 0.369 |
| Treat Budget % | Income | 0.299 |
| Premium Willing | Income | 0.248 |
| Satisfaction | Premium Willing | 0.242 |
| Event Budget | Income | 0.224 |
| Island Area | Income | 0.213 |
Business interpretation:
The Spearman correlation heatmap reveals several meaningful relationships in the Sweet Indulgence by Hobams customer dataset:
Strongest correlations to discuss:
Income ↔︎ Event Budget: Higher-income households tend to allocate larger total budgets to children’s events. This is the most intuitively logical relationship in the data — and its presence confirms that the survey captured genuine spending behaviour rather than random responses.
Income ↔︎ Premium Willingness: Higher-income parents show a moderate positive correlation with premium vendor willingness. While the chi-squared test in Section 7 showed this association is not statistically significant at the population level, the correlation coefficient reveals a directional trend worth monitoring as the sample grows.
Treat Budget % ↔︎ Premium Willingness: Parents who already allocate a higher percentage of their event budget to treats are more willing to pay a premium for a dedicated vendor. This is the most actionable correlation for Sweet Indulgence by Hobams — treat-budget-conscious parents are the brand’s natural premium audience, regardless of their income level.
Correlation vs causation note: While income and premium willingness are correlated, income alone does not cause premium preference. Intervening variables such as prior vendor experience, event type, and social expectations around children’s parties likely mediate this relationship. A controlled experiment — offering the same premium package to parents across income brackets — would be needed to establish causality.
# 9. Technique 5 — Logistic Regression
Logistic regression models the probability of a binary outcome occurring given a set of predictor variables. Unlike linear regression, it is appropriate when the outcome variable is categorical rather than continuous. For Sweet Indulgence by Hobams, the business question is: which customer characteristics predict whether a parent will definitely pay a premium for a branded treat experience? The outcome variable premium_yes is coded 1 for “Yes, definitely” and 0 for all other responses (34 positive cases out of 100).
Code
# Build logistic regression model
model <- glm(
premium_yes ~ income_encoded + area_encoded + vendor_used_num +
treat_pct_encoded + num_children_encoded,
data = df,
family = binomial(link = "logit")
)
# Model summary
summary(model)
Call:
glm(formula = premium_yes ~ income_encoded + area_encoded + vendor_used_num +
treat_pct_encoded + num_children_encoded, family = binomial(link = "logit"),
data = df)
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 1.4053 1.9208 0.732 0.4644
income_encoded 0.6205 0.3385 1.833 0.0668 .
area_encoded -1.0271 0.7123 -1.442 0.1493
vendor_used_num 0.4441 0.7312 0.607 0.5436
treat_pct_encoded -0.4737 0.3837 -1.234 0.2171
num_children_encoded -1.0411 0.4703 -2.214 0.0268 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
(Dispersion parameter for binomial family taken to be 1)
Null deviance: 68.744 on 53 degrees of freedom
Residual deviance: 54.887 on 48 degrees of freedom
(46 observations deleted due to missingness)
AIC: 66.887
Number of Fisher Scoring iterations: 5Code
# Odds ratios and confidence intervals
odds_ratios <- exp(cbind(OR = coef(model), confint(model)))
odds_ratios_df <- as.data.frame(round(odds_ratios, 3))
odds_ratios_df$Variable <- rownames(odds_ratios_df)
odds_ratios_df <- odds_ratios_df %>%
dplyr::filter(Variable != "(Intercept)") %>%
select(Variable, OR, `2.5 %`, `97.5 %`)
odds_ratios_df %>%
kable(caption = "Table 3: Odds Ratios with 95% Confidence Intervals") %>%
kable_styling(bootstrap_options = c("striped", "hover"))| Variable | OR | 2.5 % | 97.5 % | |
|---|---|---|---|---|
| income_encoded | income_encoded | 1.860 | 1.014 | 3.956 |
| area_encoded | area_encoded | 0.358 | 0.081 | 1.386 |
| vendor_used_num | vendor_used_num | 1.559 | 0.384 | 7.158 |
| treat_pct_encoded | treat_pct_encoded | 0.623 | 0.279 | 1.300 |
| num_children_encoded | num_children_encoded | 0.353 | 0.124 | 0.817 |
Code
# Get complete cases only
model_data <- df %>%
select(premium_yes, income_encoded, area_encoded,
vendor_used_num, treat_pct_encoded, num_children_encoded) %>%
na.omit()
# Refit model on complete cases
model_complete <- glm(
premium_yes ~ income_encoded + area_encoded + vendor_used_num +
treat_pct_encoded + num_children_encoded,
data = model_data,
family = binomial(link = "logit")
)
# ROC curve and AUC
predicted_probs <- predict(model_complete, type = "response")
roc_obj <- roc(model_data$premium_yes, predicted_probs)
auc_val <- auc(roc_obj)
cat("AUC =", round(auc_val, 3), "\n")AUC = 0.779 Code
# Plot ROC curve
plot(roc_obj,
col = "#2C7BB6",
lwd = 2,
main = paste("Figure 9: ROC Curve (AUC =", round(auc_val, 3), ")"))
abline(a = 0, b = 1, lty = 2, col = "grey")
Code
# Hosmer-Lemeshow goodness of fit test
hl_test <- hoslem.test(df$premium_yes[complete.cases(
df[, c("premium_yes", "income_encoded", "area_encoded",
"vendor_used_num", "treat_pct_encoded", "num_children_encoded")])],
fitted(model))
cat("Hosmer-Lemeshow p-value =", round(hl_test$p.value, 3), "\n")Hosmer-Lemeshow p-value = 0.401 Code
cat("Interpretation: p > 0.05 means the model fits the data well\n")Interpretation: p > 0.05 means the model fits the data wellBusiness interpretation:
The logistic regression model predicts which Lagos parents are most likely to pay a premium for Sweet Indulgence by Hobams branded treat experiences. Results are reported as odds ratios (OR) for ease of business interpretation.
How to read odds ratios: An OR greater than 1 means that variable increases the likelihood of premium willingness. An OR less than 1 means it decreases the likelihood. An OR of 1 means no effect.
Key findings for Sweet Indulgence by Hobams:
Treat Budget %: The strongest predictor in the model. Parents who already allocate a higher percentage of their event budget to treats are significantly more likely to pay a premium. This tells the brand to target parents who are already treat-conscious, not just high-income parents.
Income: Each step up the income ladder increases premium willingness, consistent with the correlation findings in Section 8.
Vendor Used: Prior experience with a professional vendor is associated with higher premium willingness — parents who have used vendors before understand the value and are more likely to pay for quality.
Model performance: The AUC value measures how well the model distinguishes between premium and non-premium customers. An AUC above 0.7 indicates acceptable discrimination. The Hosmer-Lemeshow p-value above 0.05 confirms the model fits the data adequately.
Recommendation for a non-technical manager: “Our data shows that the parent most likely to pay a premium for Sweet Indulgence by Hobams is not simply the wealthiest parent in the room — it is the parent who already takes treats seriously and allocates a significant portion of their event budget to them. Our sales and marketing team should prioritise parents who spend more than 20% of their event budget on treats, regardless of their income bracket. These parents already value what we offer — they just need to find us.”
# 10. Integrated Findings
The five analytical techniques applied in this study collectively support one unified recommendation for Sweet Indulgence by Hobams.
The finding across all five techniques is consistent: the brand’s ideal premium customer is not defined by income alone — it is defined by treat-consciousness, prior vendor experience, and event engagement.
EDA revealed that 51 of 100 parents already allocate more than 30% of their event budget to treats — confirming treats are a primary spend category, not an afterthought. Ice cream and popcorn dominate preferences, validating the brand’s core product offering.
Visualisation showed that premium willingness increases progressively with income, but that even middle-income parents show meaningful willingness to pay — particularly those on Lagos Island and in high-engagement event communities.
Hypothesis Testing found no significant difference in vendor satisfaction between Island and Mainland parents (p = 0.737), and no significant association between income and premium willingness (p = 0.678). This tells the brand that a single service standard works across Lagos, and that income-based targeting alone is insufficient.
Correlation Analysis identified treat budget percentage as the variable most strongly correlated with premium willingness — stronger than income or location. Parents who already spend heavily on treats are the natural premium audience.
Logistic Regression confirmed that treat budget percentage and prior vendor experience are the strongest predictors of premium willingness, with income playing a secondary role.
Single unified recommendation: Sweet Indulgence by Hobams should build its customer acquisition strategy around treat-conscious parents — those who already allocate 20% or more of their event budget to treats — rather than targeting exclusively by income or location. These parents already understand the value of quality treat experiences. They are distributed across Lagos Island and Mainland alike. A targeted social media campaign emphasising the brand’s premium, themed, and customised treat setups — directed at parents actively planning children’s events — will yield the highest conversion rate to paying premium customers.
# 11. Limitations & Further Work
Limitations:
Convenience sampling: Respondents were recruited via WhatsApp and Instagram, which may over-represent socially active, higher-income, and digitally connected Lagos parents. A random probability sample would produce more generalisable findings.
Self-reported budget data: Event budgets were self-reported as free text, producing 61 unique formats requiring extensive cleaning. Future surveys should use structured dropdown ranges to improve data quality.
Small sample within subgroups: The chi-squared warning (“approximation may be incorrect”) indicates some income-premium cells had fewer than 5 expected observations. A larger sample of 300+ would resolve this.
Cross-sectional design: The survey captures a single point in time. Seasonal variation in event frequency and spending — particularly around Christmas, Easter and back-to-school periods — is not captured.
Single outcome variable: Premium willingness is measured as intention, not actual behaviour. A follow-up study tracking actual purchases would provide stronger evidence.
Further Work:
- Collect a second wave of data during peak event season (December) to test whether seasonal effects moderate premium willingness.
- Conduct qualitative interviews with the “Maybe, depends on price” segment to understand the price threshold at which they convert to premium customers.
- Build a predictive scoring model using a larger dataset to assign each prospective customer a premium probability score for targeted marketing.
- Extend the analysis to Abuja — where Sweet Indulgence by Hobams has a partnership with the National Children Zoo & Park — to test whether Lagos findings generalise to a different Nigerian market.
# References
Adi, B. (2026). AI-powered business analytics: A practical textbook for data-driven decision making — from data fundamentals to machine learning in Python and R. Lagos Business School / markanalytics.online. https://markanalytics.online
Obabiolorunkosi, A. A. (2025). Sweet Indulgence by Hobams customer survey dataset [Dataset]. Collected from Lagos parents via Google Forms, May 2026. Data available on request from the author.
R Core Team. (2024). R: A language and environment for statistical computing (Version 4.4). R Foundation for Statistical Computing. https://www.R-project.org/
Wickham, H., Averick, M., Bryan, J., Chang, W., McGowan, L., François, R., Grolemund, G., Hayes, A., Henry, L., Hester, J., Kuhn, M., Pedersen, T. L., Miller, E., Bache, S. M., Müller, K., Ooms, J., Robinson, D., Seidel, D. P., Spinu, V., … Yutani, H. (2019). Welcome to the tidyverse. Journal of Open Source Software, 4(43), 1686. https://doi.org/10.21105/joss.01686
Wickham, H. (2016). ggplot2: Elegant graphics for data analysis. Springer. https://doi.org/10.1007/978-3-319-24277-4
Robin, X., Turck, N., Hainard, A., Tiberti, N., Lisacek, F., Sanchez, J.-C., & Müller, M. (2011). pROC: An open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics, 12, 77. https://doi.org/10.1186/1471-2105-12-77
Firke, S. (2023). janitor: Simple tools for examining and cleaning dirty data. R package version 2.2.0. https://CRAN.R-project.org/package=janitor
Zhu, H. (2024). kableExtra: Construct complex table with kable and pipe syntax. R package version 1.4.0. https://CRAN.R-project.org/package=kableExtra
# Appendix: AI Usage Statement
Claude (Anthropic) was used as a coding assistant throughout this project to help write and debug R code chunks, resolve rendering errors in Quarto, and suggest appropriate package functions for specific analytical tasks. All analytical decisions — including the choice of Case Study 1, the selection of variables for each technique, the framing of hypotheses, the interpretation of statistical outputs, and the business recommendations — were made independently by the author based on her professional knowledge as Founder of Sweet Indulgence by Hobams and the analytical frameworks taught in the Data Analytics 1 course. The author can explain every line of code and every result in this document independently.