1.1 Introduction to Descriptive Statistics
Statistics can mean numerical facts computed from data—such as averages, medians, ranges, or percentages—or the broader process of collecting, analyzing, presenting, and interpreting data. It combines principled methods with judgment in how data are collected and interpreted. Because the same data can support different interpretations, decisions should reflect the full dataset rather than an isolated number.
Data are facts and figures collected for analysis and decision-making. A data set is the complete collection for a study; an analysis may use only part of it. The basic terms describe what the data represent:
| Term | Meaning | Household survey example |
|---|---|---|
| Element | Entity about which data are collected | A household |
| Variable | Characteristic recorded for an element | Income, location, ownership, satisfaction |
| Observation | The set of measurements for one element | All recorded fields for one household; often one row, but not always |
The layout of a table does not define its variables or observations. In the apartment-rent example, 70 households each have one recorded variable—monthly rent—even though the values are displayed in seven rows and ten columns. Each rent value is an observation for one household.
Statistics supports accounting and finance analysis, economic measurement and policy, marketing and production decisions, and the monitoring of information systems.
Key takeaways
- Statistics describes numerical summaries and the process that produces and interprets them.
- An element is the entity, a variable is the characteristic, and an observation is the measurements for an element.
- Read what each value represents; do not infer data structure from table shape alone.
1.2 Types of Data
1.2.1 Types of Data I
The type of a variable determines which summaries and displays make sense. Data are broadly qualitative (categories) or quantitative (numerical amounts). A qualitative variable can still be written with numbers: numeric coding does not by itself make a category quantitative.
Qualitative scales
Nominal data are labels without a meaningful order. Ordinal data have an order, but the gaps between categories need not be measurable.
| Example | Classification | Why |
|---|---|---|
| Household location: Southwest, Northwest, Northeast, Southeast | Qualitative, nonnumeric, nominal | Names distinguish places without ranking them |
| Location coded 1–4 | Qualitative, numeric, nominal | The codes are still labels; 4 is not more or better than 1 |
| Home ownership coded 0/1 | Qualitative, numeric, nominal | Rent/own have no ranking implied by the codes |
| Satisfaction from “extremely dissatisfied” to “extremely satisfied,” or coded 1–5 | Qualitative, ordinal (nonnumeric or numeric) | The categories have an ordered ranking |
Quantitative scales
Interval measurements have meaningful, equal differences, but zero is not a meaningful absence and ratios are not meaningful. Clock time and Celsius temperature illustrate this: 4 a.m. is not twice 2 a.m.; 30°C is not three times as hot as 10°C.
Ratio measurements have meaningful differences and a meaningful zero, so ratios are meaningful. Income, profit, age, household size, price, and rent are examples: a business earning ₹2 lakh earns twice the profit of one earning ₹1 lakh, and zero profit means break-even.
1.2.2 Examples of Types of Data
Use the following classifications to distinguish categories from measured quantities:
| Data example | Qualitative variables | Quantitative variables | Scale detail |
|---|---|---|---|
| IIIT Dharwad alumni | Gender (nonnumeric nominal); domicile code (numeric nominal); rating (numeric ordinal); satisfaction label (nonnumeric ordinal) | Age, salary, work experience, overall score, core-course GPA, elective GPA | Salary, age, work experience, and an overall score with meaningful zero are ratio; GPAs on a 2–10 scale are interval. Test-score scale depends on whether zero is meaningful. |
| Canara Bank customers | None in the listed variables | Annual income, household size, monthly card spend | Ratio |
| Chroma laptop survey | Brand and operating system (nonnumeric nominal); RAM and hard disk (numeric ordinal as treated in this example) | Price and customer satisfaction score | Price and 0–100 score are treated as ratio |
| Nandini sweets weekly store data | Store (nonnumeric nominal); coupon, display, sampling indicators (numeric nominal) | Date, units sold, retail price | Date is interval; units and price are ratio |
| Mysuru apartments | Premium locality, gated complex, amenities (numeric nominal indicators) | Bathrooms, bedrooms, square footage, selling price | Ratio |
A version number such as Windows 10 or 11 is not automatically ordinal: later does not necessarily mean better. Classification depends on what the values mean, not just their appearance.
1.2.3 Types of Data II: Structure and Collection
Cross-sectional data record multiple elements at approximately one point in time. Time-series data record a variable or set of variables across multiple time periods.
| Example | Data structure | Reason |
|---|---|---|
| Household survey; alumni survey; customer card spending; laptop satisfaction; apartment prices | Cross-sectional | Different households, people, customers, or properties are observed at one survey point |
| Weekly sweets sales and promotions across stores | Time series | The same business context is observed over multiple weeks |
A survey conducted now about people who graduated at different times is still cross-sectional if the measurements are collected at one point in time. The date of the historical subject is not the same as repeated data collection over time.
Data can be obtained from existing sources (company records, government records such as the census, or specialist data organizations), observational studies (recording what occurs without controlling conditions), or experimental studies (collecting data under controlled, specified conditions for a defined target group, as in clinical trials).
Key takeaways
- Numeric codes can represent qualitative labels; determine whether values measure quantity or category.
- Nominal means unordered; ordinal means ranked; interval has meaningful differences; ratio also has meaningful zero and ratios.
- Cross-sectional versus time-series depends on when measurements are collected, not when the subject’s events occurred.
- Existing records, observation, and controlled experiments are distinct data-collection routes.
1.3 Data Visualisation
A display should fit the variable and the question. Tables and charts can summarize a single variable or show comparisons and relationships among variables. Statistical software such as Excel can generate displays; interpretation and selection matter more than the menu commands.
1.3 Data Visualisation
Categorical data: counts, proportions, and comparisons
A frequency distribution counts observations in each non-overlapping category. For the satisfaction survey, the 103 responses comprise 72 men and 31 women. Counts by satisfaction level are:
| Satisfaction | Men | Women | Overall |
|---|---|---|---|
| Very satisfied | 17 | 9 | 26 |
| Satisfied | 22 | 4 | 26 |
| Neutral | 22 | 6 | 28 |
| Unsatisfied | 9 | 7 | 16 |
| Not at all satisfied | 2 | 5 | 7 |
| Total | 72 | 31 | 103 |
The relative frequency of class is its share of the observations; percent frequency expresses that share as a percentage:
For men who are very satisfied, , or about . For women, , or about . Frequency charts use counts as bar heights; relative- or percent-frequency charts use proportions or percentages.
A pie chart divides a circle into sectors proportional to category shares. A 24% share occupies:
A sector may be labeled by count, relative frequency, or percentage. Bar charts are usually easier than pie charts for comparing category sizes because people compare lengths more accurately than sector areas. Use color, shading, and legends carefully so they clarify rather than distort.
To compare two categorical variables, use side-by-side bars or stacked bars. The arrangement depends on the comparison of interest: satisfaction levels across genders, or satisfaction composition within each gender. A stacked percentage bar chart is useful when proportions matter and group sizes do not; ordinary counts preserve differences in the number of respondents.
Quantitative data: grouped distributions and relationships
For quantitative values, a frequency distribution groups observations into non-overlapping classes. Salary classes in the example have width 5: 15–19, 20–24, 25–29, 30–34, 35–39, 40–44, and 45–50. Choose a manageable number of classes (usually five to ten), generally with equal widths, and ensure every value falls into exactly one class. Relative and percentage frequencies are computed as for categories.
A histogram plots class intervals on the horizontal axis and frequency on the vertical axis. Its adjacent rectangles touch, unlike separated bars, to show the continuous progression of quantitative intervals. The shape can reveal asymmetry, or skewness.
A scatterplot places paired observations for two quantitative variables as points; each point’s horizontal and vertical coordinates give one observation’s values. A trend line may summarize an apparent pattern. For example, plotting age against salary can reveal whether the variables appear related; the plot alone does not quantify that relationship.
| Question | Useful display |
|---|---|
| How are categories distributed? | Frequency / relative-frequency bar chart; pie chart for category shares |
| How is a quantitative variable distributed? | Histogram or grouped frequency chart |
| How do two categorical variables compare? | Side-by-side bars, stacked bars, or 100% stacked bars |
| How are two quantitative variables related? | Scatterplot, optionally with a trend line |
A data dashboard combines visual displays to help monitor organizational performance and support timely decisions.
Key takeaways
- Counts answer “how many”; relative and percent frequencies answer “what share.”
- Pie sectors encode shares as parts of ; bars generally make category comparisons clearer.
- Histograms show the shape of quantitative distributions; touching bins distinguish them from categorical bars.
- Scatterplots show paired quantitative values, while dashboards organize multiple displays for monitoring.
1.4 Measures of Central Tendency
Measures of center describe a typical or central value. The mean uses every value but is sensitive to extremes; the median is resistant; the mode identifies the most frequent value; percentiles and quartiles locate positions in an ordered distribution.
1.4 Measures of Central Tendency
Mean, median, and mode
For sample values , the sample mean is . The population mean is denoted ; the sample mean is a point estimator of it.
The median is the middle ordered value, or the average of the two middle values when is even. In , mean and median are both 2. Replacing 4 by 394 changes the mean to 80 while the median remains 2. Thus median is often more informative for income or property values with a few unusually high values.
The mode is a value with greatest frequency. One mode is unimodal; exactly two modes is bimodal; more than two is multimodal. If every value occurs once, no mode exists. In the examples and , both 1 and 2 are modes.
Percentiles and quartiles
The th percentile is a value with approximately of observations below it and above it. One common procedure sorts the values and computes the rank:
If is not an integer, linearly interpolate between the adjacent ordered values. For at the 80th percentile, , so interpolate between the 160th and 161st values. A score at the 80th percentile is approximately as high as or higher than 80% of scores. Procedures can differ slightly, especially in small samples; the Excel command discussed uses PERCENTILE.EXC.
Quartiles divide ordered data into approximately four equal groups: , (the median), and .
Salary example: alumni survey
For 103 respondents, mean salary is ₹25.9 lakh, median and mode are about ₹25 lakh. The 80th percentile is ₹27 lakh and the 90th is ₹29 lakh: approximately 80% and 90% of respondents, respectively, earned at most those amounts. Quartiles are ₹24 lakh, ₹25 lakh, and ₹27 lakh. The closeness of mean and median suggests no unusually high or low values are strongly pulling the mean, though it does not prove none exist.
Other reported summaries:
| Data | Mean | Median | Mode | ||
|---|---|---|---|---|---|
| Monthly credit-card spend, 55 customers | ₹3,852 | ₹3,864 | ₹3,657 | ₹2,890 | ₹4,531 |
| Satisfaction score, 36 customers | 57.31 | 59 | — | 53 | 61.75 |
| Weekly sweet sales, 132 observations | 135.98 | 105 | 91 | 82.25 | 136.5 |
| Property price, Mysuru | ₹202.57 lakh | ₹195 lakh | — | ₹149.5 lakh | ₹243.75 lakh |
The sweets sales mean exceeds the median substantially, consistent with some high-sales weeks; property mean exceeds median, consistent with high-priced properties lifting the average. These are clues about distribution shape, not proof of a cause.
Worked examples: bank-manager salaries and advertising spend
Bank-manager salaries. For 15 Bengaluru managers, sort the observations. The middle (8th) value is ₹8 lakh, so the sample median is ₹8 lakh. Compared with the national median of ₹8.5 lakh, the sample median is slightly lower. The salary sum divided by 15 gives a mean of ₹8.4 lakh, above the median, indicating positive skew in this sample. The percentile ranks are and , giving lakh and lakh. Roughly half the sample lies between them; roughly a quarter lies below and a quarter above .
Annual advertising spend. For 12 companies, total spend divided by 12 gives mean billion. Sort the data; the median is the average of the 6th and 7th values:
For quartiles, : interpolate 25% of the distance from billion (3rd value) to billion (4th):
Similarly, ; interpolate 75% from billion to billion:
About half of these companies’ annual advertising spend lies between billion and billion.
Weighted averages: mutual-fund returns
When group sizes differ, an overall average should weight each group’s value by an appropriate measure of its contribution. For returns and weights :
Using fund counts as weights—9,000 domestic at 5%, 2,600 international at 18%, 1,400 specialty at 12%, and 3,000 hybrid at 7%—the total count is 16,000 (not 1,600):
Fund counts may not be the most relevant weights for investor return; amounts invested are better when available. For a ₹10 lakh portfolio with ₹2 lakh domestic, ₹4 lakh international, ₹3 lakh specialty, and ₹1 lakh hybrid:
So the portfolio’s weighted return is 12.5%. The weight should represent the share of capital exposed to each return.
Key takeaways
- Mean is pulled by extremes; median is the middle ordered value; mode is most frequent.
- Percentile rank and interpolation locate percentiles under the specified convention.
- Quartiles are the 25th, 50th, and 75th percentiles; mean–median differences can signal skew.
- Weighted averages require weights that match the contribution being summarized.
1.5 Measures of Dispersion
Central values alone do not show how widely observations vary. Dispersion measures range from simple endpoint summaries to measures using every observation. Sample variance and standard deviation describe spread around the sample mean; coefficient of variation compares spread relative to the mean.
1.5 Measures of Dispersion
Range and interquartile range
The range is maximum minus minimum. It is easy to calculate but depends on only two observations, so extremes can dominate it. The interquartile range (IQR) is and measures the width of the middle half of the data, making it less sensitive to extremes.
Variance and standard deviation
Deviations from the mean sum to zero, so square them before summarizing. For a sample:
The sample variance divides the squared-deviation sum by ; the detailed reason for that denominator is deferred. The sample standard deviation is in the original units and is easier to interpret than variance. These sample statistics estimate population variance and standard deviation.
For , the mean is 2 and deviations are . Their sum is zero. Squared deviations sum to , hence:
The coefficient of variation (CV) expresses standard deviation relative to the mean:
It is a relative measure of spread. It is most interpretable when the mean is positive and the variable has a meaningful ratio scale.
Dispersion examples
| Data | Range / endpoints | Standard deviation | CV | IQR |
|---|---|---|---|---|
| Alumni salaries | — | ₹3.66 lakh | ₹27−₹24 = ₹3 lakh | |
| Credit-card spend | — | ₹1,454 | 38% | ₹4,531−₹2,890 = ₹1,641 |
| Customer satisfaction | Minimum 42, maximum 66; range 24 | 6.38 | About 11% of mean | |
| Weekly sweet sales | Minimum 34, maximum 432; range 398 | 88.36 | 65% | |
| Property prices | Minimum ₹55 lakh, maximum ₹411 lakh; range ₹356 lakh | ₹93 lakh | 46% | lakh |
Variation in sales can reflect promotional spikes rather than a problem; variability needs context. The IQR focuses on the central half, while range includes the extremes.
Worked example: smartphone use in schools
For 18 schools, the sorted usage percentages run from 14% to 77%. Their sum divided by 18 gives mean 30%. The median averages the 9th and 10th values, 26 and 28:
For quartiles, . Interpolate 75% between the 4th value (22) and 5th (23): . For , ; interpolate 25% from the 14th value (31) to the 15th (32): .
The large range relative to the IQR suggests possible outliers. Squared deviations from the mean 30 sum to . Therefore:
For the 1.5-IQR rule, . The upper fence is ; values 50 and 77 exceed it. The lower fence is ; the minimum 14 is above it. Thus 50% and 77% are candidate high outliers, with none below the lower fence.
Key takeaways
- Range uses only extremes; IQR describes the middle 50%; variance uses all squared deviations.
- Standard deviation returns to the data’s units; CV compares standard deviation to the mean.
- For the school example, range is 63 points, IQR 8.5 points, sample SD about 14.08 points, and the upper-fence candidates are 50% and 77%.
- High variability can have a business explanation; a summary alone does not establish cause.
1.6 Five Number Summaries and Box Plot
The five-number summary is minimum, , median (), , and maximum. A box-and-whisker plot displays this summary and uses the IQR to flag candidate outliers.
The box runs from to , with a line at the median. The outlier fences are:
Whiskers extend to the most extreme observed values still inside the fences; values beyond them are plotted individually as candidate outliers. A box plot may mark different observations from the -score rule.
For alumni salaries, lakh, median lakh, lakh and lakh. The lower and upper fences are ₹19.5 lakh and ₹31.5 lakh. The whiskers reach ₹21 lakh and ₹31 lakh; salaries of ₹18, ₹32, ₹35, ₹38, and ₹50 lakh are marked as outliers. The mean, ₹25.92 lakh, is shown by a cross in the illustrated plot. The -score rule flagged only ₹38 lakh and ₹50 lakh, showing that methods need not identify the same values.
A point flagged as an outlier merits investigation; the rule identifies unusual values relative to the distribution, not whether a value is erroneous.
Key takeaways
- The box spans the middle half; its line marks the median.
- Whiskers stop at the most extreme values within 1.5 IQR of the quartiles.
- Outlier flags depend on the method: here the box plot flagged five salaries, while the -score rule flagged two.
1.7 Descriptive Statistics Using Analysis Tool Pak
Excel’s Analysis ToolPak can produce multiple descriptive statistics together. Enable the Analysis ToolPak through Excel add-ins, then use Data → Data Analysis → Descriptive Statistics. Select the input range and output range, and request summary statistics. This yields a bundle of summaries—such as mean, median, variance, and other available statistics—rather than requiring each function separately.
The statistical ideas remain the same whether summaries are computed one at a time or together: central tendency, dispersion, position, and unusual observations describe collected data. Probability is introduced as the next step because it supports inference from sample data to the population, but no probability rules or worked probability examples are present in the supplied module text.
Key takeaways
- Analysis ToolPak automates a batch of descriptive summaries for a selected range.
- Descriptive statistics summarize the data observed; inference asks what the sample implies about a population.
- The provided material announces probability as a transition but ends before teaching its rules.
Additional Worked Examples: Descriptive Statistics
Bank-manager salaries
For 15 Bengaluru bank managers, sorting the salary values places ₹8 lakh in the 8th position, so the median is ₹8 lakh. The national median given for comparison is ₹8.5 lakh; the sample median is slightly lower. Adding the 15 salaries and dividing by 15 gives mean ₹8.4 lakh, above the median, suggesting positive skew.
Using , the quartile positions are:
The 4th sorted value is ₹6.7 lakh and the 12th is ₹10.6 lakh. Thus lakh and lakh; approximately half the sample lies between them.
Company advertising spend
The 12 companies’ yearly spending averages billion. With an even sample size, the median averages the 6th and 7th sorted values, billion and billion:
Quartile positions are and . Interpolating at position 3.25 between 1.97 and 2.09:
At position 9.75 between 2.56 and 2.91:
Half the companies therefore have spend between about billion and billion.
Mutual-fund weighted returns
Four categories have fund counts and returns: domestic, 9,000 and 5%; international, 2,600 and 18%; specialty, 1,400 and 12%; hybrid, 3,000 and 7%. Using counts as weights gives total weight 16,000:
The count-weighted result may not represent an investor’s return as well as invested amounts would. For a ₹10 lakh investment—₹2 lakh domestic, ₹4 lakh international, ₹3 lakh specialty, and ₹1 lakh hybrid—the investment-weighted return is:
The invested amount is the more relevant weight for this portfolio because it represents how much capital earns each category’s return.
Smartphone use across 18 schools
The ordered observations run from 14% to 77%. Their sum divided by 18 gives mean 30%. The median uses positions 9 and 10, values 26 and 28:
For , , between 22 and 23:
For , , between 31 and 32:
Thus range percentage points and IQR points. The stated sample variance is 198.12; equivalently, squared deviations from mean 30 sum to . Standard deviation is points.
The upper outlier fence is and the lower fence is . Observations 50 and 77 exceed the upper fence; no value is below 10. These two schools are candidate outliers.
Key takeaways
- Sort before locating medians and percentiles; interpolate when rank is fractional.
- A weighted return depends on the weights: fund counts yield 8.1%, while the specified investment portfolio yields 12.5%.
- In the school example, the 1.5-IQR upper fence is 44%, flagging 50% and 77% as candidates.
Additional Worked Examples: Descriptive Statistics II
Apple and Procter & Gamble monthly adjusted stock prices
Each series has 20 monthly observations. For Apple, the values sum to 1,685.43, so mean price is . For Procter & Gamble, the sum is 1,444.3, so mean is . With 20 observations, each median is the average of the 10th and 11th sorted values: Apple 83.67; Procter & Gamble 71.59. Each mean is close to its median.
For Apple, the squared deviations from the mean sum to . Dividing by gives sample variance 461.73; its square root is . Using the unrounded mean about 84.27:
For Procter & Gamble, squared deviations sum to ; thus sample variance is 20.68 and . Then:
Relative to its mean, Procter & Gamble’s observed prices vary less.
For Apple, , , so . The 1.5-IQR fences are:
The transcript reports whisker limits of 54.60 and 120.64 and no points outside the whiskers. For Procter & Gamble, , , and ; reported whisker limits are 65 and 81.43, with no points outside. The fence arithmetic and the reported whisker endpoints need not be identical: whiskers end at the most extreme observed values still inside the fences.
A scatterplot with Apple on the horizontal axis and Procter & Gamble on the vertical axis suggests that both generally rose during the period and that their prices may move together. This is an apparent positive relationship, not a quantified association or evidence of causation.
NASDAQ and Dow Jones daily percentage returns
Across 21 days in January 2021, the mean return is 0.17% for NASDAQ and −0.04% for Dow Jones. NASDAQ therefore had a marginally higher average over this period. Sample standard deviations are 1.31% and 0.9%, respectively, so NASDAQ returns varied more. Ranges are −2.61% to 2.56% for NASDAQ and −2.05% to 1.44% for Dow Jones. A scatterplot with NASDAQ on the horizontal axis and Dow Jones on the vertical axis suggests a positive relationship: days with an increase in one tended also to show an increase in the other.
Market capitalization and revenue of 14 Dow Jones companies
Market capitalization has mean billion and sample standard deviation billion:
Fiscal-year 2020 revenue has mean billion and standard deviation billion:
Relative variability is close, and slightly higher for market capitalization. The market-capitalization range is – billion. Its median is billion, billion, and billion, giving:
The reported lower and upper whisker endpoints are billion and billion; billion is a candidate outlier. Revenue ranges from billion to billion, with median billion, billion, and billion:
The reported whiskers extend from billion to billion, with no candidate outliers.
Key takeaways
- Compare variability across differently scaled series with CV: approximately 25% versus 6% for the stock-price example.
- Scatterplots can suggest co-movement; they do not establish causation.
- The NASDAQ had higher mean and variability than Dow Jones in the specified 21-day sample.
- Among the 14 companies, market capitalization has one flagged high value; the revenue box plot has none.
Test yourself on this module
Questions from these notes, one at a time, with feedback. Your weak topics are saved.