4003-01. Scientific Data & Research Methods

📊 Scientific Data & Research Methods

LO1 builds the research foundation for all engineering investigation. You will examine five quantitative and six qualitative research methodologies, apply the complete scientific method in an engineering context, and analyse data using software tools and justified graphical representations with correct SI units.

5 Quantitative methods 6 Qualitative methods Scientific method Software & graphs SI units Graphical justification

2Topics

8Examples

4Calculators

12Practice Qs

📊 This unit asks you to examine both quantitative and qualitative research methods, apply the scientific method to a real engineering problem using both types of evidence, and analyse data with software using justified graphical representations.

📊Research MethodsAll 11 methods explored

📈Data Analysis & GraphsSoftware, SI units, graphing

⚙️Applications4 data analysis tools

🎮Game12 randomised questions

📝Practice12 graded questions

📊

Quantitative, Qualitative Methods & the Scientific Method

Engineering research uses both numerical (quantitative) and descriptive (qualitative) approaches. The key insight is that both lenses investigate the same engineering problem from different angles — quantitative data establishes what is happening, qualitative data often reveals why. Every method below has a genuine engineering use case.

QuantitativeQualitativeScientific methodValidityReliability

📜 Key distinctions

Quantitative: numerical data — measurements, counts, statistics. Allows mathematical analysis, hypothesis testing, generalisation from large samples.

Qualitative: descriptive data — observations, interviews, case narratives. Explores meaning, context and experience where numerical data is insufficient.

Validity: does the method measure what it claims to measure? Internal validity (cause-effect), external validity (generalisation to other settings).

Reliability: would the same method give the same result if repeated? High reliability requires controlled, documented procedures.

Scientific method: Question → Research → Hypothesis → Experiment → Data Analysis → Conclusion & Communication. A cycle, not a checklist — conclusions feed back into new questions.

Quantitative Methods — Five methods that produce numerical data

📊 Descriptive Research Quantitative

Systematically describes and summarises the characteristics of a population, phenomenon or situation using numerical data. Does not establish cause-effect.

Engineering example: Measuring tensile strength of 200 steel samples and computing mean, SD and distribution — describes population properties without changing them.

📊 Survey Research Quantitative

Collects standardised numerical data from a sample using questionnaires, structured interviews or rating scales. Allows statistical comparison across groups.

Engineering example: Likert-scale survey of 300 workers rating machine vibration severity (1–5) — results analysed statistically to prioritise maintenance.

📊 Correlational Research Quantitative

Investigates the statistical relationship between two or more variables without manipulating them. Identifies trends but does not prove causation.

Engineering example: Plotting cutting speed vs tool wear rate over a production run — Pearson correlation coefficient r quantifies the relationship.

📊 Quasi-experimental Design Quantitative

Tests cause-effect relationships where true random assignment is not possible. Uses pre-existing groups or time-series data. More realistic than lab experiments but less controlled.

Engineering example: Comparing defect rates in two production lines before and after introducing a new lubricant — lines not randomly assigned.

📊 Experimental Research Quantitative

Manipulates an independent variable and measures the effect on a dependent variable under controlled conditions with random assignment. Highest internal validity — can establish cause-effect.

Engineering example: Testing five curing temperatures for an adhesive joint (independent) and measuring bond strength (dependent) — all other variables controlled.

Qualitative Methods — Six methods that produce descriptive data

🚫 Grounded Theory Qualitative

Develops a theory from systematically collected and analysed data — theory emerges from data rather than being imposed. Uses open, axial and selective coding of interviews and observations.

Engineering example: Interviewing maintenance engineers about fault diagnosis processes — coding responses to develop a theory of expert diagnostic reasoning.

🚫 Ethnographic Research Qualitative

Researcher immerses themselves in the natural setting of participants to understand behaviour, culture and processes from an insider perspective. Long-term observation.

Engineering example: Observing shift workers in a control room for 6 months to understand how operators make decisions under time pressure — informs system redesign.

🚫 Narrative Research Qualitative

Collects and analyses personal stories and experiences to understand how individuals make sense of events. Data: life histories, oral accounts, written narratives.

Engineering example: Recording the career narrative of experienced maintenance technicians — capturing tacit knowledge before they retire to inform training programmes.

🚫 Historical Research Qualitative

Systematically investigates past events using primary (original documents, artefacts) and secondary (literature) sources to understand causes, trends and lessons learned.

Engineering example: Analysing maintenance logs, inspection records and failure reports from a decommissioned power station to identify systemic failure patterns.

🚫 Case Study Qualitative

In-depth investigation of a single instance (one machine, one incident, one company) to understand a phenomenon in its real-world context. Rich detail, limited generalisation.

Engineering example: Detailed investigation of a single bridge collapse — examining design drawings, inspection records, material test data and witness accounts.

🚫 Phenomenology Qualitative

Explores the lived experience of participants in relation to a specific phenomenon — understanding how people perceive and make meaning of an event or situation.

Engineering example: Interviewing workers who experienced a near-miss incident — exploring their subjective experience to identify perception gaps in safety training.

The Scientific Method — Applied in engineering

1❓ Question

Identify a specific, testable engineering problem or gap in knowledge

2🔎 Research

Review existing literature; select quantitative and qualitative methods

3💡 Hypothesis

State a falsifiable prediction about the relationship between variables

4🧪 Experiment

Design and conduct controlled test; collect data using chosen methods

5📊 Data Analysis

Process data with software; apply statistics; plot justified graphs

6📢 Conclusion

Accept/reject hypothesis; communicate findings; identify limitations

📚 Key vocabulary

Independent variable: the variable the researcher deliberately changes or controls in an experiment (e.g. temperature, material type, load).
Dependent variable: the outcome measured as a result of changes to the independent variable (e.g. strength, wear rate, efficiency).
Control variable: a variable kept constant to prevent it confounding the results (e.g. humidity, specimen dimensions, operator).
Hypothesis: a falsifiable prediction stated before data collection: "If [independent variable] increases, then [dependent variable] will [increase/decrease] because..."
Triangulation: using two or more research methods (often one quantitative, one qualitative) on the same question to increase validity and cross-check findings.
Bias: systematic distortion in data collection or analysis. Quantitative: instrument bias, sampling bias. Qualitative: researcher bias, social desirability effect.

👁️ Visual aid — the scientific method: a cycle, not a linear process. Conclusions feed back to new questions.

✍️Worked example 1 — applying the scientific method in an engineering context step through it

ScenarioA manufacturer notices increased seal failures on hydraulic cylinders during winter. Apply the full scientific method to investigate whether temperature causes failure rate to increase.

1. QuestionDoes ambient temperature significantly affect hydraulic seal failure rate in this manufacturing plant?

2. Research

Review maintenance logs (historical qualitative data), manufacturer seal specifications, published correlational studies on rubber properties vs temperature.

3. Hypothesis

If ambient temperature falls below 5°C, seal failure rate will increase by more than 20% compared to temperatures above 15°C, because cold reduces seal elasticity.

4. Experiment

Quantitative: log daily failure count and temperature (12 months). Qualitative: interview maintenance technicians on observed failure patterns (ethnographic method).

5. Analysis

Correlational analysis: Pearson r between temperature and daily failures. Software (Excel/Python): scatter plot, regression line, R². Qualitative: code interview transcripts for themes.

6. Conclusion

If r < −0.7 and regression confirms 25% increase below 5°C: accept hypothesis. Recommend: heated storage for cylinders in winter; consider polyurethane seals (lower glass transition T).

Step 1 of 7

✍️Worked example 2 — comparing two research methods for the same problem

ProblemA company wants to understand why a new CNC machine has a higher-than-expected scrap rate.

Quantitative

Experimental research: systematically vary cutting speed, feed rate and depth of cut (one at a time) and measure scrap rate. Controlled; statistical analysis; identifies optimal parameters.

Qualitative

Case study: observe operators over two weeks, review all scrap tickets, interview setters. Reveals that night-shift operators skip warm-up cycle — a human factors issue not captured by numbers alone.

Triangulation

Combining both: experimental data sets optimal parameters (quantitative) AND operator behaviour study identifies training gap (qualitative). Triangulation gives a complete picture.

AnswerQuantitative optimises parameters. Qualitative uncovers human/process factors. Together they triangulate to a complete solution. ✓

✍️Worked example 3 — grounded theory in practice

QuestionDescribe how grounded theory would be used to investigate why maintenance engineers in a petrochemical plant consistently underestimate repair times.

Data collectionSemi-structured interviews with 15 engineers. Field observations during repairs. Analysis of work-order completion data.

Coding

Open coding: identify concepts (optimism bias, tool availability, hidden complexity). Axial coding: link categories. Selective coding: core category = "optimism under uncertainty".

Theory

Theory emerges: engineers underestimate because they anchor on best-case scenarios and lack feedback on past estimates due to poor record-keeping systems.

AnswerGrounded theory produces a data-driven theory of underestimation — actionable for management (improve records, introduce pre-task time analysis). ✓

✍️Worked example 4 — choosing and justifying a method

QuestionA structural engineer wants to determine whether a new carbon-fibre composite layup produces higher fatigue life than a standard layup. Identify the most appropriate research method and justify it.

Method

Experimental research (quantitative): independent variable = layup type (standard vs new); dependent variable = cycles to failure; controlled: specimen dimensions, loading frequency, amplitude, temperature.

Why not correlational?

Correlational only observes — it cannot confirm that the layup causes the difference. Experimental research allows cause-effect conclusion.

Why not case study?

Case study is qualitative, single-instance — no statistical power to confirm general superiority of one layup over another.

AnswerExperimental research — only method with sufficient control to establish a causal relationship between layup type and fatigue life. ✓

🔨 Try it yourself

Now you can try this calculator with your own values.

📊Research method explorerSelect any method for full analysis

Method type

Select any research method to explore its purpose, procedure and engineering application.

📈

Data Analysis, Software & Graphical Presentation

Analysing engineering data requires choosing the right software, the right chart type and the right statistical measures — then justifying every decision. Graphs must have correct SI units, a title, a scale and a best-fit line where appropriate. The key habit: every graph needs 2–3 sentences explaining what it shows and why that chart type was the right choice for the data.

SI unitsy = mx + clog-logsemi-logMean / SD / SEError propagation

📜 Key rules — data analysis

Mean: x̄ = Σx ÷ n. Sample SD: s = √[Σ(x−x̄)² ÷ (n−1)]. Standard error: SE = s ÷ √n. SE measures reliability of the mean.

Propagation — add/subtract: Δ(a±b) = Δa + Δb. Multiply/divide: δ(ab) = δa + δb. Power: δ(aⁿ) = |n|δa. (δ = relative/fractional uncertainty)

Graph choice: scatter + best-fit line for continuous relationships; bar chart for categorical comparisons; histogram for distributions; pie chart only for parts-of-a-whole.

Linearisation — power law y=Axⁿ: plot log y vs log x. Exponential y=Ae𝑾𝑥: plot ln y vs x. Straight line means model is valid.

SI units to know: Pa (N/m²), J (N·m), W (J/s), Hz (s⁻¹), N (kg·m/s²), T (Wb/m²), V (W/A), Ω (V/A).

📚 Key vocabulary — data and graphs

SI unit: Système International d'Unités — the internationally agreed standard units. All engineering data must be expressed in SI units on graphs and in reports.
Best-fit line: a straight line drawn so data points are balanced above and below it — determined by least squares. Always drawn through the general trend, not point-to-point.
Error bar: plotted as ± uncertainty on each data point. If the best-fit line passes through all error bars, the model is consistent with the data.
Linearisation: applying a mathematical transform (log, ln) to a non-linear equation so it plots as a straight line — enabling gradient and intercept to be read accurately.
Coefficient of determination R²: how well a model fits data. R² = 1 is perfect; R² < 0.8 suggests a poor fit or wrong model. Calculated by regression software (Excel, Python, MATLAB).
Anomalous result: a data point that lies significantly off the trend. Must be identified, investigated (measurement error? equipment fault?) and excluded with justification if confirmed invalid.

👁️ Visual aid — accuracy vs precision (left) and the effect of sample size on standard error: SE = s/√n (right)

✍️Worked example 5 — mean, SD & SE from repeat measurements step through it

QuestionFive repeat measurements of a shaft diameter (mm): 24.3, 24.7, 24.5, 24.6, 24.4. Calculate the mean, sample standard deviation and standard error. All measurements in mm.

Meanx̄ = (24.3+24.7+24.5+24.6+24.4) ÷ 5 = 122.5 ÷ 5 = 24.5 mm

Deviations²(−0.2)²+(0.2)²+(0)²+(0.1)²+(−0.1)² = 0.04+0.04+0+0.01+0.01 = 0.10

SDs = √(0.10 ÷ 4) = √0.025 = 0.158 mm

SESE = 0.158 ÷ √5 = 0.158 ÷ 2.236 = 0.0707 mm

ReportDiameter = 24.5 ± 0.07 mm (standard error, n=5). State the unit (mm) on all reported values.

Answerx̄ = 24.5 mm s = 0.158 mm SE = 0.071 mm ✓

Step 1 of 7

✍️Worked example 6 — justifying a graph type

QuestionAn engineer records load (kN) vs deflection (mm) for a beam under 8 different loads. State which graph type to use, justify it, and describe what the gradient means.

Graph type

Scatter plot with straight best-fit line (least squares). Axes: x = load (kN), y = deflection (mm). Both continuous variables with a linear relationship expected from Hooke's law.

Why not bar chart?

Bar charts compare discrete categories. Load and deflection are continuous — a scatter plot preserves the relationship between values and allows the gradient to be extracted.

Gradient

gradient = Δdeflection / Δload (mm/kN) = flexibility (inverse of stiffness). k = 1/gradient (kN/mm). R² confirms linearity.

Software

Excel: Insert → Scatter → Add trendline → Display equation and R². Python: matplotlib.scatter + numpy.polyfit. Both produce gradient and intercept automatically.

AnswerScatter + best-fit line, justified by continuous variables and linear model. Gradient = flexibility; R² confirms fit. ✓

✍️Worked example 7 — error propagation in a formula

QuestionPower P = IV. I = 2.50 ± 0.05 A, V = 12.0 ± 0.2 V. Find P and its percentage uncertainty.

PP = 2.50 × 12.0 = 30.0 W

Relative errorsδI = 0.05/2.50 = 0.020 (2.0%) δV = 0.2/12.0 = 0.01667 (1.667%)

Combined%δP = 2.0 + 1.667 = 3.667% → ΔP = 0.03667 × 30.0 = 1.10 W

AnswerP = 30.0 ± 1.1 W (3.7%)

✍️Worked example 8 — log-log linearisation

QuestionA spring force law gives F = kxⁿ. A log-log plot of F vs x has gradient 1.03 and passes through (log 5, log 24.7). Find k and n. State the SI units.

nn = gradient = 1.03 (close to 1 — nearly linear spring)

log klog k = log F − n×log x = log(24.7) − 1.03×log(5) = 1.3927 − 0.7195 = 0.6732

kk = 10^0.6732 = 4.71 N/mⁿ (SI unit: N/m^1.03 for this non-integer exponent)

Why odd units?Since F = kxⁿ, rearranging gives k = F/xⁿ, so [k] = N/mⁿ. With n = 1.03 (not a clean integer), k's units come out as N/m^1.03 — not a standard named unit, but dimensionally honest. If the spring were perfectly linear (F = kx), the fitted gradient would land exactly on n = 1.00 and k would have the tidy units N/m. Getting 1.03 instead is a signal that the real spring is slightly non-linear, or that there is experimental scatter nudging the fitted line's gradient off 1.

AnswerF = 4.71x^1.03 n = 1.03, k = 4.71. Units: F in N, x in m. ✓

✍️Worked example 9 — a full investigation, start to finish step through it

ScenarioA robotic welding cell is producing inconsistent weld penetration depth on a batch of steel brackets, and the rejection rate has crept up to 9% with no obvious single cause.

Experimental research

Run the cell at five welding current settings (180–220 A in 10 A steps), producing 10 test welds at each. Sample and measure penetration depth on a cross-section micrometer. Independent variable: current; dependent variable: penetration depth; all other settings held constant.

Graph for this data

Scatter plot with best-fit line: welding current (x, A) vs penetration depth (y, mm). Justified — both variables continuous, and a roughly linear relationship is expected over this range. Gradient gives mm of extra penetration per amp.

Case study

Alongside the trial, one operator's shift is observed in detail over a full week: torch angle is logged at intervals, and every rejected weld's defect photo is reviewed. Rich, single-cell detail rather than a large sample — appropriate here because the problem may be local to this one cell, not general to the process.

Graph for this data

Bar chart of defect-type counts logged during the case-study week (porosity, undercut, incomplete penetration, spatter). Justified — defect type is a discrete category, not a continuous scale, so a bar chart compares frequency across groups; a scatter or line graph would not apply here.

What each analysis showsThe experimental data quantifies a real, controllable effect: penetration depth rises predictably with current, so current alone can explain some of the variation. But the case-study observation adds something the numbers can't — it shows torch angle drifting late in each shift, correlating with a spike in undercut and incomplete-penetration defects, most likely caused by a loosening fixture clamp rather than the welding parameters at all.

ConclusionRecommend setting current to the value from the experimental trial that gives reliable penetration, and separately scheduling a mid-shift fixture-clamp check to remove the drift the case study identified. Numerical and non-numerical evidence together point to two distinct, addressable causes that neither method would have found alone. ✓

Step 1 of 7

👁️ Visual aid — the two graphs from Worked Example 9, side by side

🔨 Try it yourself

Now you can try this calculator with your own values.

📊Statistics calculator — mean, SD & standard errorEnter comma-separated values

Data values (comma separated)

Enter values above to compute statistics.

⚙️

Engineering Applications

Four data analysis tools supporting research methods, statistics, error propagation and graphical analysis.

📊 Research method selector

Enter your engineering investigation scenario and identify the most appropriate research method type.

Quantitative: numerical data, hypothesis testing. Qualitative: context, meaning, experience.

🧮 Method advisor

What do you want to find out?

Select your research goal above.

📊 Statistics calculator

Enter comma-separated values to get mean, SD, SE and percentage uncertainty in the mean.

x̄ = Σx/n s = √[Σ(x−x̄)²/(n−1)] SE = s/√n

🧮 Calculator

Values (comma separated)

Enter data values.

📈 Error propagation calculator

Find the percentage uncertainty in Z = AⁿB𝑚 given uncertainties in A and B.

%δZ = |n|×%δA + |m|×%δB

🧮 Calculator

Power of A (n)

%uncertainty in A

Power of B (m)

%uncertainty in B

Enter powers and percentage uncertainties.

🔌 Graph type justifier

Select your data type and relationship to get a justified graph recommendation with software guidance.

Match data type to graph type — and justify the choice

🧮 Advisor

Data type

Select your data type.

🎮

Research Methods & Data Challenge

12 randomised questions — methods, statistics, graphing and the scientific method.

📊 Quick-fire round

Question 1 / 12Score: 0Streak: 0 🔥

Press Start to begin!

📝

Practice Questions

12 questions covering research methods, the scientific method, statistics, error propagation and graphical analysis.

📊 For qualitative answers, write in complete engineering sentences. For numerical answers, show full working and include units.

Q1 · Quantitative methods2 marks

Name three quantitative research methods from the Unit 4003 specification and state one key characteristic of each.

Any three of: Descriptive (summarises numerical data without intervention), Survey (collects numerical ratings from a sample), Correlational (measures statistical relationship between variables, no manipulation), Quasi-experimental (tests cause-effect without random assignment), Experimental (manipulates independent variable under controlled conditions).

Q2 · Qualitative methods2 marks

Name three qualitative research methods from the Unit 4003 specification. State what type of data each produces.

Any three of: Grounded theory (coded interview transcripts → emergent theory), Ethnographic (field observation notes → cultural/behavioural understanding), Narrative (personal stories → lived experience data), Historical (document analysis → pattern identification over time), Case study (multi-source investigation of one instance), Phenomenology (structured interviews → subjective experience themes).

Q3 · Quantitative vs qualitative2 marks

An engineer needs to investigate why a new assembly robot has a 12% higher reject rate than the old one. State one quantitative and one qualitative method that should be used and justify each choice.

Quantitative: Experimental — vary one robot parameter at a time (speed, grip force, alignment) and measure reject rate. Justified: identifies which parameter causes the fault.
Qualitative: Case study or Ethnographic — observe operator-robot interaction over a week; review all reject tickets. Justified: reveals process, setup or training issues that data alone cannot show.

Q4 · Scientific method3 marks

State all six stages of the scientific method in the correct order.

1. Question → 2. Research → 3. Hypothesis → 4. Experiment → 5. Data Analysis → 6. Conclusion and Communication

Q5 · Scientific method in context4 marks

A new heat treatment process is claimed to increase yield strength of an aluminium alloy by 15%. Apply the scientific method (all 6 stages) to this investigation. Include one quantitative and one qualitative method.

1. Question: Does the new heat treatment increase yield strength by ≥15%?
2. Research: Review literature on age-hardening of aluminium; identify current baseline strength data (historical method).
3. Hypothesis: If the new treatment is applied, yield strength will increase by ≥15% compared to standard treatment, because precipitation hardening increases dislocation resistance.
4. Experiment: Experimental research — 20 specimens per condition; tensile test per BS EN ISO 6892-1; measure yield strength (0.2% proof stress).
5. Analysis: Calculate mean, SD, SE for each group; t-test for statistical significance; bar chart of means with error bars; Excel analysis.
6. Conclusion: Accept/reject hypothesis; report with SI units (MPa); state limitations (batch variation, operator effects).

Q6 · Mean and SD2 marks

Stress readings (MPa): 210, 215, 212, 208, 215. Calculate the mean and sample standard deviation.

x̄ = (210+215+212+208+215)/5 = 1060/5 = 212 MPa
Deviations: −2, 3, 0, −4, 3. Squared: 4, 9, 0, 16, 9. Sum=38.
s = √(38/4) = √9.5 = 3.08 MPa

Q7 · Standard error2 marks

Using Q6 data (s = 3.08 MPa, n = 5), calculate the standard error and state what it tells you about the reliability of the mean.

SE = s/√n = 3.08/√5 = 3.08/2.236 = 1.377 MPa
Interpretation: the true population mean is estimated to lie within ±1.38 MPa of our sample mean with 68% confidence. A smaller SE would require more measurements.

Q8 · Error propagation2 marks

Kinetic energy E = ½mv². m = 5.0 ± 0.1 kg, v = 10.0 ± 0.2 m/s. Find %uncertainty in E.

δm = 0.1/5.0 = 0.02 (2.0%) δv = 0.2/10.0 = 0.02 (2.0%)
%δE = 2.0 + 2×2.0 = 6.0%
E = ½×5×100 = 250 J ΔE = 0.06×250 = 15 J E = 250 ± 15 J

Q9 · Graph justification2 marks

An experiment measures temperature (°C) vs resistance (Ω) for a thermistor. The relationship is non-linear. State which graph type you would use and how you would linearise the data.

Thermistor: R = Ae^(B/T). Linearise: plot ln R vs 1/T (semi-log variant). Alternatively: plot ln R vs T if approximately exponential. This gives a straight line with slope = B (a material constant) and intercept = ln A. Justify: straight line confirms model validity; gradient and intercept can be read accurately.

Q10 · Software analysis2 marks

Name two software tools used to analyse engineering data and state one statistical output each can produce.

Excel: regression trendline with equation and R²; AVERAGE, STDEV, NORM functions; chart with best-fit line and error bars.
Python (matplotlib/numpy/scipy): polyfit for regression; pearsonr for correlation; scatter plots with error bars. MATLAB, Minitab or LabVIEW also acceptable with justification.

Q11 · SI units2 marks

State the SI units for: (a) pressure, (b) energy, (c) frequency, (d) magnetic flux density, (e) electrical resistance.

(a) Pressure: Pa (N/m²)
(b) Energy: J (N·m = kg·m²/s²)
(c) Frequency: Hz (s⁻¹)
(d) Magnetic flux density: T (Wb/m²)
(e) Electrical resistance: Ω (V/A)

Q12 · Combined investigation5 marks

A manufacturing company wants to reduce energy consumption in their injection moulding process. Design a full research investigation: name one quantitative and one qualitative method (with justification), apply the scientific method, and specify the graph and software you would use to present findings.

Quantitative: Experimental research — vary mould temperature and cycle time; measure energy consumption (kWh) per part. Justified: controlled IV/DV, statistical comparison.
Qualitative: Ethnographic — observe operators over 4 weeks; note informal practices (machine left idle, suboptimal settings). Justified: reveals human factors missed by instruments.
Scientific method: 1) Question: Can mould temperature optimisation reduce energy by ≥10%? 2) Research: Review literature and process data. 3) Hypothesis: Reducing temperature from 80°C to 65°C will reduce energy by ≥10% without reducing part quality. 4) Experiment: 50 parts per temperature; measure kWh and defect rate. 5) Analysis: Excel scatter plot of temperature (x) vs energy (y); regression line; R²; error bars showing SD. 6) Conclusion: Report mean energy savings (kWh, SI unit) with uncertainty; recommend optimal temperature.
Software: Excel for regression and chart. Python/Minitab for ANOVA if multiple conditions compared. Every graph should be followed by 2–3 sentences explaining what it shows and why that chart type was chosen.