English:Data Science Foundations
Data Science Foundations
Introduction
Data Science Foundations introduces you to the ideas and habits used to turn data into evidence. The course is designed for Grades 9–10 and connects Mathematics, Statistics, Computer science, Data visualization, and responsible decision-making. You do not need advanced programming. You need curiosity, careful reasoning, basic arithmetic, and a willingness to ask whether a conclusion is really supported by the data.
Data science is not simply "finding numbers." It is a process of asking a useful question, collecting or finding appropriate data, checking its quality, exploring patterns, building explanations or simple models, communicating results, and thinking about limits and ethical consequences. A strong data scientist also knows when the evidence is too weak to support a claim.
The diagram above shows why data science is interdisciplinary. Statistical thinking helps you reason under uncertainty, computing helps you work efficiently with data, and subject knowledge helps you interpret what the numbers mean in context.
By the end of this aiMOOC, you should be able to:
- Ask a measurable data question: Turn a broad topic into a question that can be investigated with data.
- Collect or locate data responsibly: Distinguish populations, samples, variables, and possible sources of bias.
- Check and clean a data set: Find missing values, duplicates, inconsistent labels, impossible values, and unit problems.
- Summarize data: Use measures such as mean, median, range, and interquartile range appropriately.
- Choose and interpret visualizations: Use charts that fit the variable types and avoid misleading displays.
- Reason about relationships: Interpret scatterplots and correlation without confusing association with causation.
- Use simple prediction ideas: Understand what a fitted line does and why predictions have uncertainty.
- Evaluate ethical issues: Consider consent, privacy, fairness, representation, and responsible communication.
The video introduces statistics as a set of tools for learning from data. As you watch, notice the distinction between using data to support a decision and treating statistics as automatic truth.
What Is Data Science?
Data science is the study and practice of extracting useful knowledge from data. It combines methods from statistics, computing, and the field where the data came from. A sports analyst, climate researcher, health researcher, journalist, business analyst, and school administrator may all use data science, but they ask different questions and must understand different contexts.
A useful way to think about data science is as a cycle:
Question → Data → Quality check → Exploration → Model or explanation → Communication → Decision or new question
The cycle matters because the quality of the final conclusion depends on every earlier step. A sophisticated model cannot repair a badly framed question or a biased sample. Likewise, a correct calculation can still produce a misleading conclusion if the data do not represent the group you care about.
The workflow diagram illustrates a broader data process: information is collected, cleaned, explored, modeled, and turned into a product or decision. Real projects often move backward as well as forward. For example, an unusual chart may reveal a data-entry error, sending you back to the cleaning stage.
Data, Information, and Evidence
Data are recorded observations. They can be numbers, categories, text, images, locations, timestamps, sensor readings, or other measurements. Data become useful when you connect them to a question and interpret them in context.
Information is organized or summarized data. A table of daily temperatures becomes more informative when you calculate a weekly average or plot a line chart.
Evidence is information used to support or challenge a claim. Evidence is stronger when the data are relevant, collected fairly, measured reliably, analyzed appropriately, and presented with important limitations.
Suppose a class wants to know whether students who sleep longer tend to score higher on a quiz. A spreadsheet containing sleep hours and quiz scores is data. A scatterplot showing the relationship is information. A carefully stated conclusion such as "students reporting more sleep tended to have higher quiz scores in this class, but the data do not prove that extra sleep caused the difference" is an evidence-based interpretation.
Asking Good Data Questions
A strong data project begins with a question that is specific, measurable, and connected to available data. "Is school stressful?" is broad. "How does self-reported homework time relate to self-reported stress ratings among students in our grade this week?" is more measurable.
Before collecting anything, identify:
- Population: The full group you want to understand.
- Sample: The smaller group you actually observe.
- Variable: A characteristic recorded for each case.
- Unit of observation: What one row represents, such as one student, one day, or one city.
- Operational definition: The exact rule used to measure an idea.
An operational definition prevents vague concepts from changing during a project. If you study "screen time," decide whether that means total device time, entertainment only, or time outside school. If you study "fitness," decide whether it means resting heart rate, a timed run, or another measure.
A useful question should also be realistic. Ask whether you can collect the data safely, ethically, and accurately. Sensitive personal topics may not be appropriate for a school project even if they are interesting.
From Question to Variables
Imagine this question: Do students who spend more time studying tend to earn higher quiz scores?
The unit of observation could be one student. The explanatory variable could be study time in minutes. The response variable could be quiz score as a percentage. You would also need to think about other factors, such as previous knowledge, difficulty of the quiz, sleep, attendance, or access to study resources.
This step matters because data science is about relationships between clearly defined measurements, not just collecting as many columns as possible.
Understanding Data and Variables
A data set is an organized collection of cases and variables. In a simple table, each row usually represents one case and each column represents one variable.
| Student ID | Study minutes | Quiz score | Sleep hours | Study location |
|---|---|---|---|---|
| A01 | 35 | 72 | 7.5 | Home |
| A02 | 60 | 88 | 8.0 | Library |
| A03 | 20 | 65 | 6.5 | Home |
| A04 | 50 | 84 | 7.0 | Library |
The table contains different variable types.
Categorical variables place observations into groups. "Study location" is categorical because its values describe categories such as Home or Library.
Numerical variables represent quantities where arithmetic is meaningful. Study minutes, quiz score, and sleep hours are numerical.
Some categorical variables have a meaningful order, such as low, medium, and high. These are often called ordinal variables. Other categories, such as eye color, have no natural ranking.
Some numerical variables are counts, such as number of books read. Others are measurements, such as height or temperature. Recognizing the variable type helps you choose suitable summaries and charts.
Units and Metadata
Numbers without units can be dangerous. A height value of 170 could mean centimeters, inches, or something else. A temperature of 30 means something different in degrees Celsius and degrees Fahrenheit.
Metadata are data about data. Useful metadata can include variable names, units, definitions, collection dates, missing-value codes, sources, and measurement methods. Good metadata make a data set easier to understand and less likely to be misused.
Collecting Data and Sampling
You often cannot measure an entire population, so you collect a sample. The goal is not merely to get many responses. The goal is to get observations that are relevant to the question and as representative as reasonably possible.
A random sample uses a chance process so members of the population have a known opportunity to be selected. A convenience sample uses people who are easy to reach. Convenience samples are faster, but they can create selection bias.
Suppose you want to estimate average travel time to school for all Grade 9 students. Surveying only students who arrive very early might miss students with longer or less reliable journeys. The sample would then systematically differ from the population of interest.
The sampling video explains why survey wording and the way participants are selected can change results. A large biased sample can still produce a misleading estimate.
Bias, Randomness, and Measurement Error
Bias is a systematic tendency that pushes results in a particular direction. Common sources include:
- Selection bias: Some groups are more likely to enter the sample than others.
- Response bias: People may answer inaccurately because of wording, pressure, memory, or social expectations.
- Measurement error: A tool or procedure may not measure the intended variable accurately.
- Nonresponse bias: People who do not respond may differ from people who do.
Random variation is different from bias. Even a fair random sample will not match the population perfectly. If you take several random samples, the results will vary. Statistical reasoning helps you separate normal sampling variation from patterns that may reflect real differences.
Data Quality and Cleaning
Raw data are rarely ready for analysis. Data cleaning means identifying and handling problems so that later calculations are meaningful.
Common issues include missing values, duplicate rows, inconsistent spelling, impossible values, mixed units, incorrect data types, and unexpected outliers. Cleaning should be documented rather than hidden.
Consider these entries for a variable called "height_cm":
| Raw value | Possible issue | Responsible action |
|---|---|---|
| 172 | None obvious | Keep after checking units |
| 1.68 | Mixed unit or scale | Verify whether this means meters |
| -165 | Impossible sign | Check the original record |
| blank | Missing value | Mark as missing and investigate |
| 900 | Extreme or erroneous | Verify before deciding what to do |
Do not automatically delete values simply because they look unusual. An outlier can be a mistake, but it can also be a real and important observation. The correct action depends on the context and the evidence available.
A Reproducible Cleaning Log
A good cleaning log records what changed and why. For example: "Converted all heights from meters to centimeters using the documented unit field" is better than "fixed heights." Reproducibility means another person can understand and repeat your process.
When possible, keep the original raw data unchanged and create a cleaned copy. This protects the evidence trail and lets you undo a mistaken decision.
Describing a Distribution
A distribution describes how values are spread across a numerical variable. To understand a distribution, look at its center, spread, shape, and unusual values.
The mean is the sum of the values divided by the number of values. It uses every value and is useful when the distribution is reasonably balanced and extreme values are not dominating the result.
The median is the middle value after sorting the data. It is often more resistant to extreme values.
The range is maximum minus minimum. It is simple but depends only on two observations.
The interquartile range describes the spread of the middle half of the data. It is useful when the median is a suitable measure of center.
The video reviews mean, median, and mode. For data-science work, the important question is not just whether you can calculate a statistic, but whether that statistic is appropriate for the distribution and the question.
Histograms and Shape
A histogram groups numerical values into intervals called bins and shows how many observations fall into each interval.
When reading a histogram, ask:
- Where is the center?
- How spread out are the values?
- Is the shape roughly symmetric or skewed?
- Are there gaps, clusters, or possible outliers?
- Would a different bin width change the story?
A histogram is not the same as a bar chart. Histogram bars represent intervals of a numerical scale, so adjacent bars normally touch. Bar charts compare categories, so the spaces between bars help signal that the categories are separate.
Normal Distributions as a Reference Shape
Some numerical data are approximately bell-shaped and symmetric. A mathematical model called the normal distribution is often used as a reference for this kind of pattern.
Not every real-world variable is normally distributed. Income, waiting times, online views, and many other variables can be strongly skewed. You should inspect the data rather than assume a particular shape.
Visualizing Data
A good visualization helps the reader see a pattern accurately. The best chart depends on the question and variable types.
| Goal | Useful display | What to look for |
|---|---|---|
| Compare categories | Bar chart | Differences in counts or percentages |
| Show one numerical distribution | Histogram or box plot | Center, spread, shape, outliers |
| Compare two numerical variables | Scatterplot | Direction, form, strength, unusual points |
| Show change over time | Line chart | Trends, cycles, sudden changes |
A chart should have a clear title, labeled axes, units, and an honest scale. Avoid decorative effects that make values harder to compare.
Misleading graphs may truncate an axis to exaggerate differences, use unequal intervals, hide relevant groups, or choose only a time window that supports a preferred story. Data literacy includes checking how a graph was designed, not just reading its headline.
Data Visualization as Reasoning
Visualization is not only presentation. It is also an analysis tool. A scatterplot may reveal a nonlinear pattern that a single correlation value hides. A histogram may reveal two clusters that a mean alone cannot show.
A useful workflow is: plot early, calculate next, explain last. Looking at the data can prevent you from summarizing the wrong feature.
Relationships Between Variables
A scatterplot displays pairs of numerical values. Each point represents one case. If higher values of one variable tend to occur with higher values of another, the relationship is positive. If higher values of one tend to occur with lower values of the other, the relationship is negative.
Correlation measures the direction and strength of a linear relationship between two numerical variables. A correlation near +1 indicates a strong positive linear relationship. A correlation near -1 indicates a strong negative linear relationship. A correlation near 0 indicates little linear relationship, but a curved relationship may still exist.
The central lesson is crucial: correlation does not prove causation. Two variables can move together because one affects the other, because the direction runs the opposite way, because a third variable affects both, or because the apparent pattern is partly due to chance or selection.
Confounding and Alternative Explanations
A confounding variable is related to both the explanatory variable and the outcome, making causal interpretation difficult.
Imagine that students who carry water bottles also tend to score higher on a fitness test. It would be unreasonable to conclude that owning a water bottle causes fitness. Students who exercise more may be both more likely to carry water and more likely to perform well.
Good data reasoning asks: What else could explain this pattern? This question is often more valuable than a quick causal claim.
From Relationships to Prediction
A model is a simplified representation of a pattern. In a scatterplot with an approximately straight relationship, a linear regression line can summarize the trend and produce predictions.
A regression line does not pass through every point. Instead, it is fitted to represent the overall relationship. Predictions close to the center of the observed data are usually more defensible than predictions far outside the observed range.
For Grades 9–10, focus on the conceptual idea: a fitted line gives a predicted value, while the vertical differences between observed points and predicted values are called residuals. Smaller residuals indicate closer predictions for those cases.
Training and Testing a Model
When a model is built and evaluated on exactly the same data, it can appear better than it really is. A basic machine-learning habit is to separate data into a training set used to fit the model and a test set used to evaluate it on unseen cases.
This idea protects against overfitting, where a model follows accidental details of the training data instead of learning a pattern that generalizes.
A model can also fail because the future population differs from the training data. Data science therefore requires continued checking, not a one-time accuracy score.
A First Look at Machine Learning
Machine learning uses algorithms that learn patterns from data. It is part of data science, but data science is broader: many valuable projects use tables, visualizations, and statistical reasoning without machine learning.
In supervised learning, the training data include a target or label to predict. Predicting a numerical value, such as tomorrow's temperature, is a regression task. Predicting a category, such as whether an email is spam or not spam, is a classification task.
In unsupervised learning, there is no target label. The goal may be to discover structure, such as grouping similar observations into clusters.
The Iris scatterplot shows measurements from three iris species. The example is widely used to demonstrate how measured features can form patterns useful for classification.
The comparison above hints at an important idea: machine-learning systems can combine multiple decision rules or models. At this level, you do not need to master random forests. Instead, notice that increased model complexity can make interpretation harder, so accuracy is not the only goal.
Features, Labels, and Evaluation
A feature is an input variable used by a model. A label or target is what the model tries to predict.
Evaluation should match the task. For a classification problem, overall accuracy can be useful, but it may hide poor performance for a smaller group. For a numerical prediction problem, you can examine how far predictions are from actual values.
Always compare performance with a reasonable baseline. A complicated model that barely improves on a simple rule may not be worth the added complexity.
Data Ethics, Privacy, and Fairness
Data science affects people. Ethical practice asks not only "Can we analyze this data?" but also "Should we?" and "Who could be helped or harmed?"
Important principles include:
- Informed consent: People should understand what data are being collected and how they may be used when consent is required.
- Privacy: Personal information should be protected, and unnecessary identifying details should not be collected.
- Data minimization: Collect only what is needed for the legitimate purpose.
- Fairness: Check whether data and models create systematically different outcomes for groups.
- Transparency: Explain important methods, assumptions, limits, and uncertainty.
- Security: Protect data from unauthorized access or accidental exposure.
The history of research and data collection includes serious ethical failures. Modern data work must respect human dignity, consent, privacy, and the possibility that technical choices can have unequal consequences.
Bias in Data and Models
Bias can enter before a model is built. Historical records may reflect unequal opportunities. A training sample may underrepresent some groups. A measurement may work differently in different contexts. A target variable may be only a rough proxy for the real concept of interest.
A model trained on biased data can reproduce or amplify those patterns. That is why fairness cannot be checked only at the final step.
For a school project, use low-risk topics when possible. Avoid collecting unnecessary names, contact details, health information, precise locations, or other sensitive personal information. Ask a teacher or responsible adult about school rules before collecting data from people.
Communicating a Data Story
A strong data story combines a clear question, relevant evidence, an appropriate visual, and a careful conclusion. It does not hide uncertainty.
A useful structure is:
- Question: What did you want to learn?
- Data: Where did the data come from, and who or what is represented?
- Method: What cleaning, summaries, visualizations, or models did you use?
- Finding: What pattern did you observe?
- Limitation: What could weaken the conclusion?
- Next step: What new data or analysis would improve the investigation?
Use plain language. Instead of saying "The data proves students learn better in the morning," say "In this sample, morning classes had higher average scores. The data do not establish that time of day caused the difference."
Uncertainty Is Part of the Result
Data science does not remove uncertainty. It organizes uncertainty so that decisions can be better informed.
A responsible conclusion distinguishes what you observed from what you infer. It also separates predictions from guarantees. If the data set is small, biased, old, or measured imprecisely, say so.
This habit is central to scientific thinking and responsible communication.
Mini Project: From Question to Conclusion
Suppose your class asks: How is travel time to school related to arrival time?
You could define one row as one school day for one student, then record travel time in minutes and arrival time relative to the first lesson. Before collecting data, decide whether names are necessary. In most classroom projects, anonymous or coded records are safer.
After collection, check for missing entries, impossible negative travel times, inconsistent units, and duplicate submissions. Make a histogram of travel time and a scatterplot of travel time against arrival time. Calculate a suitable measure of center for travel time. If the scatterplot shows a trend, describe it without claiming causation.
Your final conclusion might be: "Longer travel times were associated with later arrival in this sample, but the relationship varied substantially. The sample came from one class and one week, so it may not represent the whole school or other seasons."
That statement includes a finding and a limitation. It is more useful than a dramatic claim that ignores the data-collection context.
Data Science Checklist
Before accepting a conclusion, ask:
- Data question: Is the question clear and measurable?
- Data source: Do the data actually represent the group or process of interest?
- Data quality: Were missing values, errors, units, and duplicates checked?
- Statistical summary: Are the summaries appropriate for the distribution?
- Data visualization: Does the chart show the data honestly?
- Correlation and causation: Is an association being mistaken for a cause?
- Model evaluation: Was prediction performance checked on data not used to fit the model?
- Data ethics: Were privacy, consent, fairness, and possible harms considered?
- Scientific communication: Are uncertainty and limitations stated clearly?
Interactive Tasks
Quiz: Test Your Knowledge
What best describes data science? (Using data, statistics, computing, and context to answer questions) (!Collecting as many numbers as possible without a question) (!Making charts without checking data quality) (!Using machine learning for every problem)
Which variable is numerical? (Height in centimeters) (!Favorite school subject) (!Type of transport) (!Eye color)
Which measure of center is usually more resistant to one extreme outlier? (Median) (!Mean) (!Range) (!Maximum)
What is a main purpose of random sampling? (To reduce systematic selection bias) (!To guarantee every result is correct) (!To remove all random variation) (!To make every sample identical)
Which graph is most suitable for showing the distribution of one numerical variable? (Histogram) (!Pie chart) (!Flowchart) (!Venn diagram)
What can a correlation between two variables establish by itself? (A statistical association) (!A guaranteed causal relationship) (!The direction of every individual case) (!That no third variable matters)
What does a simple linear regression line mainly provide? (A summary of a linear trend and predicted values) (!Proof that one variable causes another) (!A way to remove every outlier) (!A guarantee of perfect future predictions)
Which action belongs to responsible data cleaning? (Checking inconsistent units before analysis) (!Deleting every unusual value automatically) (!Changing values until the pattern looks clear) (!Replacing missing data without documentation)
Why is a test set useful in predictive modeling? (It checks performance on data not used to fit the model) (!It makes the training data unnecessary) (!It guarantees the model will be fair) (!It removes the need to inspect errors)
Which practice best supports ethical data collection? (Collect only necessary data and protect privacy) (!Collect personal details in case they become useful) (!Hide the purpose of the project from participants) (!Publish identifiable records for transparency)
Memory Game
| Mean | Sum of values divided by the number of values |
| Median | Middle value after the data are ordered |
| Outlier | Observation far from most other values |
| Sample | Observed subset of a larger population |
| Variable | Characteristic recorded for each case |
| Scatterplot | Graph that displays pairs of numerical values |
| Correlation | Measure of the direction and strength of a linear association |
| Residual | Difference between an observed value and its prediction |
Drag and Drop
| Match the correct terms. | Topic |
|---|---|
| Collect | Gather observations that are relevant to the question |
| Clean | Check missing values, units, duplicates, and errors |
| Explore | Use summaries and visualizations to search for patterns |
| Model | Represent a relationship or make a prediction |
| Communicate | Explain findings, uncertainty, and limitations |
...
Crossword Puzzle
| Dataset | What organized collection usually contains cases and variables? |
| Median | What statistic is the middle value after ordering data? |
| Outlier | What do you call an observation far from most others? |
| Sampling | What process selects observations from a population? |
| Scatterplot | What graph displays pairs of numerical values as points? |
| Regression | What method can fit a line for numerical prediction? |
LearningApps
Cloze Text
Open-Ended Tasks
Easy
- Data diary: For one day, record ten non-sensitive examples of data you encounter, such as temperatures, prices, times, or scores, and label each example as categorical or numerical.
- Question makeover: Rewrite four vague questions about school life as specific, measurable data questions and identify the population, unit of observation, and variables for each.
- Mini survey: Design a five-question anonymous survey on a low-risk school topic, test it with two classmates, and revise any wording that could cause confusion or response bias.
- Chart critique: Find or photograph a chart from a newspaper, textbook, poster, or public website and create an annotated image showing what the chart does well and what could mislead a reader.
Standard
- Data cleaning lab: Create a small spreadsheet containing deliberate errors such as duplicates, mixed units, missing values, and inconsistent labels, exchange it with a partner, and produce a cleaning log that explains every correction.
- School data story: Collect or use an approved class data set, create at least two suitable visualizations, and write a one-page data story with a question, finding, limitation, and next step.
- Sampling investigation: Run a simple experiment comparing a convenience sample with a random sample on the same harmless question, then explain how the sampling method changed the result.
- Correlation hunt: Find a public data set with two numerical variables, make a scatterplot, describe the relationship, propose at least two alternative explanations, and record a short video explaining why correlation alone does not prove causation.
Advanced
- Prediction project: Build a simple linear prediction from a small approved data set, separate training and test cases, compare predictions with actual values, and explain where the model performs poorly.
- Bias audit: Choose a hypothetical school decision system, such as predicting course support needs, and produce a fairness audit identifying possible sampling bias, measurement bias, privacy risks, and safer design choices.
- Data dashboard: Design a one-page dashboard from an open public data set with three coordinated charts, concise explanatory text, source information, and a section explaining what the dashboard cannot tell the reader.
- Interview a data professional: Interview a data analyst, researcher, librarian, journalist, or other professional who works with data, or visit a relevant workplace, university lab, library, or science center, then create a podcast or video report comparing their workflow with the process taught in this course.
Learning Assessment
- Question design assessment: Given a broad claim such as "technology improves learning," develop a measurable research question, define the population and variables, and justify what evidence would and would not address the claim.
- Sampling and bias assessment: Compare two proposed sampling plans for a school survey, identify likely sources of bias in each, and defend which plan would produce more credible evidence.
- Distribution assessment: Given a numerical data set with one extreme value, compare mean and median, choose the more informative summary, and explain how the outlier changes your interpretation.
- Visualization assessment: Examine two different charts made from the same data, determine which communicates the pattern more honestly, and justify your answer using scale, labels, chart type, and context.
- Causation assessment: Evaluate a claim based on a strong correlation, identify at least two alternative explanations or confounding variables, and describe what additional evidence would strengthen a causal argument.
- Model transfer assessment: A prediction model works well on last year's students but poorly this year; propose several data or context changes that could explain the drop and recommend a fair way to reevaluate the model.
- Ethics assessment: Review a proposed project that collects identifiable student data, identify privacy and fairness risks, and redesign the project using data minimization, clearer consent, and safer reporting.
Evidence of Learning
| Area | Evidence you can provide |
|---|---|
| Knowledge | You can explain populations, samples, variables, distributions, correlation, regression, bias, and basic machine-learning terms in your own words. |
| Data skills | You can organize a small data set, identify quality problems, document cleaning decisions, calculate suitable summaries, and create appropriate charts. |
| Reasoning | You can distinguish description from inference, association from causation, and prediction from certainty. |
| Ethical judgment | You can identify privacy, consent, representation, fairness, and security concerns and suggest safer alternatives. |
| Products | You can produce a cleaned data table, visualization, short analysis, presentation, dashboard, image annotation, podcast, or video that communicates evidence clearly. |
| Transfer | You can apply the workflow to a new school, science, social, environmental, or community question and explain how the context changes the method. |
OERs on the Topic
Useful related open learning topics include Statistics, Descriptive statistics, Data visualization, Sampling, Correlation, Linear regression, Machine learning, Data ethics, and Scientific method.
Linked Learning Areas
