Food Analysis Lab · For teachers

Learning Assessment

Students use different tools (Excel, Orange, Colab), so assessment should look at judgment and conclusions, not software skills. This page is a ready-to-use assessment design: pre/post tests show how much was learned over the semester, weekly exit tickets and homework feedback show where students get stuck, and practical exams and the project show whether they can independently analyze data they have not seen.

Download the assessment toolkit (Excel) Data analysis tools guide Semester plan

Four principles

Align with goalsEach learning goal needs at least two different kinds of evidence, and assessment items map directly to goals.

Tool-neutralOne rubric scores work in all three tools; students who get it right in Excel earn full marks, and Colab is only a bonus.

Formative first, summative laterWeekly feedback is ungraded or low-stakes so students dare to make mistakes; grading is concentrated on reports, practical exams, and the project.

Data-informed teachingPre/post tests and item analysis reveal concepts the whole class has not mastered, to adjust next semester.

Learning goals and evidence

Goal	Students can…	Main evidence
CLO1	Correctly carry out proximate analysis and complete the calculations (wet/dry basis, %N × factor, recovery, dilution factor)	Lab reports, weekly data homework, midterm practical, pre/post tests
CLO2	Use statistics to judge data quality and draw conclusions (SD, CI, tests, calibration curve, z-score, control chart, uncertainty)	Weekly data homework, midterm practical, pre/post tests
CLO3	Build and evaluate spectroscopic chemometric models (preprocessing, PCA, PLS, RMSECV/RMSEP, applicability domain)	Final practical, sweetener project report card, pre/post tests
CLO4	Choose a suitable tool for the problem and verify the result with a second tool	“Cross-check” item in the practical exams, self-efficacy survey item 9
CLO5	Communicate data clearly in writing and orally	Lab reports, W17 presentation and peer review

Assessment plan for the semester

Assessment	Time	How	Type
Concept pre/post test	W1, W18	15 multiple-choice questions (toolkit sheet “2_Pre/post item bank”); the same test before and after, 15 minutes, using Google Forms quiz mode.	Ungraded
Self-efficacy survey	W1, W9, W18	10 items rated 1–5 (e.g., “I can tell whether a result is accurate and whether it is precise”). Compare with the pre/post test: high confidence with a wrong answer signals a misconception.	Ungraded
Exit ticket	Last 3 minutes of each week	1 concept question + “the one thing I understood least today.” The following week, spend 5 minutes at the start responding to the point most students did not understand.	Formative
Weekly data homework	W1–W8, W11–W16	Complete the practice questions in Excel, Orange, or Colab (any one); score with the rubric and add one specific suggestion to each.	Formative 15%
Lab report	Every lab	Scored on five rubric dimensions: calculations, quality-control judgment, figures, conclusions and error, records.	Summative 35%
Midterm practical	W9	Given unseen method-comparison or proficiency-test data, complete a quality-control report in Excel or Colab (your choice).	Summative 15% (includes written exam)
Final practical	W18	Given a set of spectra and reference values, complete PCA and PLS in Orange or Colab (your choice) and write the model conclusions.	Summative 15% (includes written exam)
Sweetener project	W15–W17	Model report card (LV, RMSECV, RMSEP, RPD, applicability domain) + oral presentation + peer review.	Summative 20%
Advanced bonus	Whole semester	Complete advanced questions in Colab/Jupyter (e.g., VIP, T² and Q, NNLS), or cross-check the same result with two tools and explain the difference.	Up to +5

Lab reports 35%

Weekly homework 15%

Midterm 15%

Final 15%

Sweetener project 20%

Uses the weights from the semester plan, only changing “weekly Colab homework” to any tool. Pre/post tests and surveys are ungraded so students answer honestly.

Rubric: score judgment, not software

Dimension (weight)	4 Excellent	3 Good	2 Needs work	1 Insufficient
Correct calculations (25%)	Values, units, and significant figures all correct; matches the answer-check table	1 minor error that does not affect conclusions	Errors affect some conclusions	Main values wrong or missing
Quality-control judgment (25%)	Every judgment cites data and a threshold	Mostly correct; a few lack support	Conclusions without support, or wrong method	No judgment
Figures (15%)	Suitable plot type; axes, units, error bars complete	Some labels missing	Unsuitable plot or unclear error bars	No figure
Conclusions and error discussion (25%)	Separates systematic and random error, identifies the largest source, and proposes improvements	Discusses but does not distinguish, or proposes no improvement	Only “human error”	No discussion
Records and reproducibility (10%)	Raw data, calibration records, and files attached; others can recalculate	Complete but unclear naming	Missing raw data or calibration records	Cannot be traced

Total score = Σ(level × weight) ÷ 4. Toolkit sheet “8_Rubrics” has a calculator and also the practical-exam rubric (with one extra item, “cross-check”: verify 1 value a second way). Give the rubric to students in the first week.

How to set the practical exams

W9 midterm: method comparison or proficiency test

Give an unseen long-format dataset (e.g., manual vs. automatic acetic acid titration for 6 groups, or blind-sample ash for 5 groups).
Task: descriptive statistics → find outliers → choose the right test (paired t / Bland–Altman or ANOVA) → write a 3-sentence conclusion.
Grading focus: choose the right method and give the reason, correct values, and a conclusion that answers the question.
Tools: Excel (open book; students may use their own EX04, EX08) or Colab.

W18 final: spectral modeling

Give a set of spectra and reference values, with the calibration and test sets already labeled.
Task: preprocessing → PCA score plot and interpret one cluster → choose LV for PLS → report RMSECV, RMSEP, RPD → state the applicability domain.
Grading focus: grouped cross-validation, drawing conclusions from RMSEP rather than R², and naming the model’s limitations.
Tools: Orange (students may adapt OR13, OR16) or Colab.

You can prepare two versions of the data (Version A and B) so neighbors get different ones; work out the answers beforehand in the Colab notebooks and grade against the toolkit rubric.

How to tell whether students improved

1. Normalized gain g

g = (post − pre) ÷ (max − pre). For the class: <g> = (post mean − pre mean) ÷ (max − pre mean)

It means “how much of what was not yet known got learned.” Hake (1998) levels: g < 0.3 low, 0.3–0.7 medium, > 0.7 high. Compare the class <g> with previous years to see whether teaching changes worked; also check the paired t-test p-value and effect size (Cohen's d) to confirm the improvement is not by chance.

2. Item analysis

Proportion correct P: items still below 0.5 on the post-test are concepts the class has not mastered; strengthen that week next semester.
Discrimination r (correlation of the item score with the total score): items below 0.2 may be unclear and need rewriting.
Summarize by learning goal (CLO1–CLO3) to see which strand is weakest.

3. Confidence vs. ability

High self-efficacy but a wrong answer on the matching test item means students “think they know it”; these misconceptions most need direct attention in class.

4. End-of-semester review

After the semester, spend 30 minutes with toolkit sheets “5_Learning gains” and “6_Item analysis” and write down 3 things to adjust next semester (e.g., “add one more residual-plot exercise in W7”), then add them to the semester plan.

Sample pre/post test questions

The full 15 questions (covering W1–W16 topics) are in toolkit sheet “2_Pre/post item bank”, with answers and linked goals, ready to paste into Google Forms. Questions test concepts only, so students can answer them whichever tool they use.

See 5 sample questions

1. (W1) Repeating the same pipette 10 times gives a very small SD, but the mean is 0.03 mL below the nominal value (Class A tolerance ±0.02 mL). This means?
Precise but not accurate / Accurate and precise / Accurate but not precise / Neither accurate nor precise

2. (W2/W4) Two methods measure the same set of samples. To compare the methods, you should use?
Paired t-test or Bland–Altman plot / Independent two-sample t-test / Look only at the correlation coefficient r / Compare only the two SDs

3. (W7) A calibration curve has R² = 0.999. What does this mean?
The residual plot may still be curved; check the residuals too / It is definitely linear / LOD is definitely low / Sample concentrations are definitely accurate

4. (W8) The specification is protein ≥ 25.0%, and the measured value is 25.2 ± 0.4% (U, k = 2). What should be concluded?
Cannot decide: the interval crosses the specification limit / Release / Do not release / Measure once more and average, then release

5. (W13) In cross-validation, the 3 replicates of the same sample are split between the training and validation sets. RMSECV will be?
Too low (over-optimistic) / Too high / Unchanged / Impossible to calculate

What is in the assessment toolkit

Sheet	Purpose
1_Goals and evidence	Learning goals × assessment map (editable)
2_Pre/post item bank	15 concept questions, answers, linked weeks and goals
3_Pre-test answers, 4_Post-test answers	Paste students’ A/B/C/D answers for automatic scoring (up to 60 students)
5_Learning gains	Individual and class normalized gain, paired t-test, effect size, pre/post bar chart
6_Item analysis	Per-item proportion correct, improvement, and discrimination, summarized by goal, with weak items flagged automatically
7_Self-efficacy survey	10-item survey and W1/W9/W18 mean comparison chart
8_Rubrics	Homework/report rubric and practical-exam rubric, with a scoring calculator
9_Grade summary	Calculates semester grades by weight, including bonus points and letter grades

The student data in the toolkit are simulated demo data; clear the yellow student-data area before real use.