experimental-data.qmd

---
title: "Working with experimental data in R"
subtitle: "Tables, transformations, and comparisons"
author: "Denis O'Meally"
institute: "City of Hope"
format:
  revealjs:
    theme:
      - default
      - brand
    slide-number: c/t
    transition: none
    controls: true
    progress: true
    hash: true
    history: true
    center: false
    width: 1600
    height: 900
    margin: 0.08
    code-overflow: wrap
    embed-resources: true
    footer: "BIOSCI 504 · Working with experimental data in R"
execute:
  warning: false
  message: false
---

## By the end

```{r}
#| label: setup
#| include: false

library(BIOSCI504)
library(tidyverse)

data(mouse_trial, package = "BIOSCI504")
data(course_data_dictionary, package = "BIOSCI504")

illustrative_comparison <- tibble(
  comparison_group = rep(c("A", "B"), each = 5),
  outcome = c(1, 2, 3, 4, 5, 2.8, 2.9, 3, 3.1, 3.2)
)

illustrative_outcome_plot <- ggplot(
  illustrative_comparison,
  aes(x = comparison_group, y = outcome, color = comparison_group)
) +
  scale_color_manual(values = c(A = "#177BFB", B = "#0757A8")) +
  labs(
    x = "Illustrative group",
    y = "Illustrative outcome",
    caption = "Illustration only — not mouse_trial results"
  ) +
  theme_minimal(base_size = 24)

paired_measurements <- tribble(
  ~mouse, ~measurement, ~weight_g,
  "Mouse A", "Baseline", 20,
  "Mouse A", "Final", 25,
  "Mouse B", "Baseline", 24,
  "Mouse B", "Final", 25
) |>
  mutate(measurement = factor(measurement, levels = c("Baseline", "Final")))

paired_measurement_plot <- ggplot(
  paired_measurements,
  aes(x = measurement, y = weight_g, group = mouse, color = mouse)
) +
  geom_line(linewidth = 2) +
  geom_point(size = 6) +
  scale_color_manual(values = c("Mouse A" = "#177BFB", "Mouse B" = "#0757A8")) +
  scale_y_continuous(limits = c(18, 27), breaks = seq(18, 26, by = 2)) +
  labs(x = NULL, y = "Weight (g)", color = NULL) +
  theme_minimal(base_size = 24) +
  theme(legend.position = "bottom")
```

You should be able to:

1. state what one row represents;
2. inspect a table before changing it;
3. choose an operation that matches the biological question;
4. inspect a derived variable before comparing it;
5. distinguish a result, its interpretation and the evidence supporting it.

::: {.notes}
- Purpose: prepare the reasoning used in the exercise without completing it.
- Ask: If every line runs, which outcome is already established?
- Answer: none; execution is not verification.
- Build source: `BIOSCI504` commit `3730f8b813668f317689ac56fabb299c9a1b072e`.
:::

## How this session works

```text
LOOK → PREDICT → TRY IN RSTUDIO → DISCUSS → REVEAL
```

- The slides contain the code and precomputed output.
- Short pauses are marked **Now in RStudio**.
- Work with your neighbour; the TAs will circulate around the room.

::: {.notes}
- You present from this deck; no instructor RStudio window is needed.
- Students use RStudio only at the marked pauses.
- Advance to the following slide for the precomputed answer.
:::

## The mouse experiment

:::: {.columns}

::: {.column width="29%"}
### Before

**48 mice**

Baseline weight
:::

::: {.column width="6%"}
### →
:::

::: {.column width="30%"}
### Assigned group

Control

Nutritional treatment
:::

::: {.column width="6%"}
### →
:::

::: {.column width="29%"}
### After

Final weight
:::

::::

`sex` · `cage` · `batch` provide experimental context for each mouse.

**Did treated mice gain more weight than control mice?**

Treatment varies within cages; treatment and cage are not the same grouping.

::: {.notes}
- Context: 48 mice, one assigned group, weights before and after treatment.
- Clarify: treatment varies within cages; cage is context, not the treatment group.
- Ask: What information must the table retain to answer the question?
- Do not discuss the observed treatment result.
:::

## Before seeing the table

Predict:

- What should one row represent?
- Which columns are needed?
- Which column should identify a mouse?
- Where should the unit of weight be recorded?

::: {.notes}
- Take answers before advancing.
- Listen for mouse ID, treatment, baseline weight and final weight.
- Accept sex, cage and batch as useful experimental context.
:::

## Before seeing the table

Predict:

- What should one row represent?
- Which columns are needed?
- Which column should identify a mouse?
- Where should the unit of weight be recorded?

::: {.callout-note appearance="simple" icon=false}

| One row | Identifier | Assigned group | Measurements | Context |
|---|---|---|---|---|
| one mouse | `mouse_id` | `treatment` | baseline and final weight | sex, cage, batch |

One value per cell. Units belong in metadata or the column name.
:::

::: {.notes}
- Reveal this as a defensible answer, not the only imaginable layout.
- Takeaway: table structure should preserve experimental structure.
- Transition: inspect the actual representation.
:::

## Four rows from the real table {.smaller}

```{r}
#| echo: false

mouse_trial |>
  slice_head(n = 4) |>
  select(
    mouse_id,
    treatment,
    sex,
    baseline_weight_g,
    final_weight_g,
    cage,
    batch
  ) |>
  knitr::kable()
```

::: {.notes}
- This is a genuine preview from `mouse_trial`.
- Ask learners to locate the identifier, group, measurements and context.
- A preview does not establish completeness, uniqueness or plausible ranges.
:::

## What does one row represent?

Choose one:

1. one weight measurement;
2. one mouse;
3. one cage;
4. one treatment group.

::: {.notes}
- Ask for a show of hands before advancing.
- Do not resolve the choice on this slide.
:::

## What does one row represent?

Choose one:

1. one weight measurement;
2. one mouse;
3. one cage;
4. one treatment group.

::: {.callout-note appearance="simple" icon=false}

**One mouse.**

The two measurements occupy different columns on the same mouse row.
:::

::: {.notes}
- Answer: one mouse.
- Takeaway: identifiers, duplicates, sample size and change depend on this choice.
- Return to the observational unit after transformations and summaries.
:::

## Rows, columns and cells

| Part | Meaning in this table |
|---|---|
| row | one observed mouse |
| column | one recorded or derived variable |
| cell | one value for one mouse and variable |
| `mouse_id` | the identifier for the observational unit |

One value per cell; units belong in metadata or the column name.

::: {.notes}
- Connect the vocabulary directly to the displayed table.
- Takeaway: rows are observations, columns are variables, cells are values.
- Source: https://datacarpentry.github.io/spreadsheet-ecology-lesson/01-format-data.html
:::

## Variables have different roles

| Role | Variables |
|---|---|
| identifier | `mouse_id` |
| assigned group | `treatment` |
| biological/context variables | `sex`, `cage`, `batch` |
| continuous measurements | `baseline_weight_g`, `final_weight_g` |

The role follows from the experiment, not merely the R storage type.

::: {.notes}
- Connect to Lecture 2: character and double describe storage.
- Treatment, sex, cage and batch are categorical variables stored as text.
- Weight is a continuous measurement.
:::

## The dictionary is part of the data {.smaller}

```{r}
#| echo: false

course_data_dictionary |>
  filter(dataset == "mouse_trial") |>
  transmute(
    variable,
    description,
    units = replace_na(units, "—"),
    allowed_values = replace_na(allowed_values, "—")
  ) |>
  knitr::kable()
```

::: {.notes}
- A name alone is not a complete variable definition.
- Ask: could the R type tell us whether `27500` is a plausible weight?
- Answer: no; the dictionary supplies units and biological meaning.
:::

## Checkpoint: read the representation

For the cell in row 2 under `final_weight_g`:

- What does its row represent?
- What does its column represent?
- What does the value represent?
- Where does its unit come from?

::: {.notes}
- Take answers before advancing.
- Check that row and measurement are no longer being conflated.
:::

## Checkpoint: read the representation

For the cell in row 2 under `final_weight_g`:

- What does its row represent?
- What does its column represent?
- What does the value represent?
- Where does its unit come from?

::: {.callout-note appearance="simple" icon=false}

- **Row:** mouse M002.
- **Column:** post-treatment weight.
- **Cell:** M002's recorded final weight.
- **Unit:** grams, established by `_g` and the data dictionary.
:::

::: {.notes}
- Reveal and correct any mismatch in the room's answers.
- Takeaway: meaning comes from table structure plus metadata.
:::

## Now in RStudio

Run once in the Console:

```r
library(tidyverse)
data(mouse_trial, package = "BIOSCI504")
```

Keep `mouse_trial` unchanged. We will build temporary pipelines from it.

::: {.notes}
- Pause while everyone loads the package data.
- The TAs circulate; confirm that `mouse_trial` appears in the Environment.
- The next slides contain precomputed output if anyone falls behind.
:::

## A preview is not an inspection

```r
head(mouse_trial)
```

can show examples of values.

It cannot establish:

- the total number of records;
- whether every ID is unique;
- missingness elsewhere;
- complete group counts;
- the full range of a measurement.

::: {.notes}
- Ask: what could be wrong in row 40 that `head()` cannot reveal?
- Takeaway: a preview samples rows; inspection checks the complete object.
- Source: https://datacarpentry.github.io/R-ecology-lesson/instructor/how-r-thinks-about-data.html
:::

## Inspection starts with expectations

| Question | Check | Concerning result |
|---|---|---|
| Is the table complete? | dimensions | missing or extra records |
| Does one row equal one mouse? | distinct IDs | duplicates |
| Are measurements present? | missingness | incomplete data |
| Are groups represented? | counts | imbalance or miscoding |
| Are values plausible? | ranges + units | impossible or mis-scaled values |

Inspection is a sequence of questions, not one command.

::: {.notes}
- Ask learners to state an expected result before each check.
- Takeaway: a check is useful when a failure would change the analysis.
:::

## Now in RStudio: inspect structure

Predict what each command will tell you, then run it:

```r
dim(mouse_trial)
names(mouse_trial)
glimpse(mouse_trial)
```

::: {.notes}
- Allow two minutes; pairs can divide the three commands.
- Ask for dimensions and one observation about the schema.
- Advance for the precomputed output; do not open instructor RStudio.
:::

## Now in RStudio: inspect structure {.smaller}

Predict what each command will tell you, then run it:

```{r}
#| echo: true

dim(mouse_trial)
names(mouse_trial)
glimpse(mouse_trial)
```

::: {.notes}
- Answer: 48 rows and seven columns.
- Point out character identifiers/groups and double weight measurements.
- `glimpse()` does not test every data-quality condition.
:::

## Check the observational unit

```r
mouse_trial |>
  summarise(
    rows = n(),
    distinct_mice = n_distinct(mouse_id)
  )
```

What relationship do you expect between the two results?

::: {.notes}
- Ask for the expected relationship before students run it.
- Allow one minute, then advance for the precomputed answer.
:::

## The identifiers support the row claim

```{r}
#| echo: true

mouse_trial |>
  summarise(
    rows = n(),
    distinct_mice = n_distinct(mouse_id)
  )
```

Rows equal distinct mouse IDs in this table.

::: {.notes}
- Answer: both are 48.
- A mismatch would trigger investigation; it would not identify the cause.
- Do not introduce the challenge fixture here.
:::

## Now in RStudio: groups and missingness

Before running, predict the number of groups and what `0` would mean:

```r
mouse_trial |>
  count(treatment)

mouse_trial |>
  summarise(
    across(
      c(baseline_weight_g, final_weight_g),
      ~ sum(is.na(.x))
    )
  )
```

::: {.notes}
- Take predictions, then allow two minutes to run both checks.
- Advance for the precomputed output.
:::

## Now in RStudio: groups and missingness

Before running, predict the number of groups and what `0` would mean:

```{r}
#| echo: true

mouse_trial |>
  count(treatment)

mouse_trial |>
  summarise(
    across(
      c(baseline_weight_g, final_weight_g),
      ~ sum(is.na(.x))
    )
  )
```

::: {.notes}
- Group counts establish denominators before comparison.
- There is one missing baseline weight and one missing final weight.
- These need not occur in the same mouse; the derived change has two missing values.
- Later checks may examine missingness within groups.
:::

## Now in RStudio: check ranges against units

Predict a plausible weight range in grams, then run:

```r
mouse_trial |>
  summarise(
    across(
      ends_with("_weight_g"),
      ~ list(range(.x, na.rm = TRUE))
    )
  )
```

The suffix `_g` makes the expected scale testable.

::: {.notes}
- Ask for a rough plausible range before execution.
- Advance for the precomputed output.
:::

## Now in RStudio: check ranges against units

Predict a plausible weight range in grams, then run:

```{r}
#| echo: true

mouse_trial |>
  summarise(
    across(
      ends_with("_weight_g"),
      ~ list(range(.x, na.rm = TRUE))
    )
  )
```

::: {.notes}
- Compare the output with the dictionary and biological expectation.
- A valid double can still use the wrong unit.
- Takeaway: plausible storage is not plausible biology.
:::

## Checkpoint: choose the revealing check

Which check is most direct?

| Concern | Candidate check |
|---|---|
| the same mouse appears twice | ? |
| one weight uses another unit | ? |
| final weights are missing mainly in one group | ? |

::: {.notes}
- Ask pairs to supply one check for each concern.
- Require them to name the evidence each check would produce.
:::

## Checkpoint: choose the revealing check

Which check is most direct?

| Concern | Candidate check |
|---|---|
| the same mouse appears twice | ? |
| one weight uses another unit | ? |
| final weights are missing mainly in one group | ? |

::: {.callout-note appearance="simple" icon=false}

- **Duplicate mouse:** compare rows with distinct IDs, then inspect duplicates.
- **Wrong unit:** compare ranges or plots with the documented unit.
- **Group-dependent missingness:** count missing values within treatment.
:::

::: {.notes}
- Reveal these as defensible checks, not magic commands.
- Takeaway: each concern should map to evidence capable of exposing it.
:::

## The question determines the operation

:::: {.columns}

::: {.column width="48%"}
### Final weight

Are treated mice heavier at the end?
:::

::: {.column width="48%"}
### Weight change

Did treated mice gain more weight?
:::

::::

These are different biological questions.

::: {.notes}
- This is the conceptual centrepiece.
- Final weight combines starting weight and change during the study.
- Neither target is inherently invalid; the operation must match the question.
:::

## Same final weight, different change

```{r}
#| echo: false
#| fig-height: 5.2
#| fig-alt: "Two mice both finish at 25 grams. Mouse A rises from 20 to 25 grams, while Mouse B rises from 24 to 25 grams."

paired_measurement_plot
```

Mouse A gained **5 g**. Mouse B gained **1 g**. Both finished at **25 g**.

::: {.notes}
- Ask: which quantity answers the study question?
- Takeaway: paired measurements retain information the final value alone cannot.
- This illustration does not reveal the treatment comparison.
:::

## Valid code can answer the wrong question

Question: **Did treated mice gain more weight?**

```r
mouse_trial |>
  group_by(treatment) |>
  summarise(
    n = sum(!is.na(final_weight_g)),
    mean_final_g = mean(final_weight_g, na.rm = TRUE)
  )
```

This code runs. Which question does it actually answer?

::: {.notes}
- Ask learners to name the mismatch before advancing.
- The code is valid and its output could look reasonable.
- Do not run or reveal the treatment means.
:::

## Valid code can answer the wrong question

Question: **Did treated mice gain more weight?**

```r
mouse_trial |>
  group_by(treatment) |>
  summarise(
    n = sum(!is.na(final_weight_g)),
    mean_final_g = mean(final_weight_g, na.rm = TRUE)
  )
```

This code runs. Which question does it actually answer?

::: {.callout-note appearance="simple" icon=false}

It compares **mean final weight** between treatment groups—not weight gain.

“It ran” does not establish that the code implemented the scientific question.
:::

::: {.notes}
- Answer: it asks whether groups differ in final weight.
- Takeaway: compare the specification with the executed operation.
:::

## Name the quantity before coding

For each mouse:

$$
\text{weight change (g)} = \text{final weight (g)} - \text{baseline weight (g)}
$$

Expected sign: positive for gain, negative for loss.

::: {.notes}
- Move from words to arithmetic to an R name.
- Units remain grams because grams are subtracted from grams.
- Ask for the expected sign before showing the transformation.
:::

## Now in RStudio: derive the variable

```r
mouse_trial |>
  mutate(
    weight_change_g = ____________________
  )
```

What should remain unchanged after `mutate()`?

::: {.notes}
- Give pairs one minute to complete the expression.
- Answer: `final_weight_g - baseline_weight_g`.
- Invariant: still 48 rows, each representing one mouse.
:::

## The derived table state {.smaller}

```{r}
#| echo: true

mouse_trial |>
  mutate(
    weight_change_g = final_weight_g - baseline_weight_g
  ) |>
  select(mouse_id, baseline_weight_g, final_weight_g, weight_change_g) |>
  slice_head(n = 6) |>
  knitr::kable()
```

`mutate()` adds a column; it does not change what one row represents.

::: {.notes}
- Ask students to verify one subtraction from the displayed values.
- Takeaway: inspect the new table state before continuing.
- Source: https://datacarpentry.github.io/R-ecology-lesson/instructor/working-with-data.html#making-new-columns-with-mutate
:::

## Inspect the new state

```{r}
#| echo: true

mouse_trial |>
  mutate(
    weight_change_g = final_weight_g - baseline_weight_g
  ) |>
  summarise(
    type = typeof(weight_change_g),
    missing = sum(is.na(weight_change_g)),
    minimum = min(weight_change_g, na.rm = TRUE),
    maximum = max(weight_change_g, na.rm = TRUE)
  )
```

::: {.notes}
- Ask what each check contributes: type, missingness and plausible range.
- Stop before grouping the real changes by treatment.
- Takeaway: inspect → transform → inspect again.
:::

## Missing inputs propagate

```text
final weight     baseline weight     change
    25.3 g    −       NA          =    NA
      NA      −      24.2 g       =    NA
```

The mouse remains a row. The derived measurement is missing.

::: {.notes}
- Connect to Lecture 2: arithmetic cannot reconstruct an unobserved input.
- Do not silently delete these rows.
- Record the effective sample and whether missingness differs by group.
:::

## Checkpoint: predict the transformation

For one mouse:

```text
baseline = 24.1 g
final    = 25.3 g
```

Before calculating:

- What sign should the change have?
- What is its value and unit?
- What does the row represent after adding the column?

::: {.notes}
- Take answers before advancing.
- Use this as a small-case test of the transformation.
:::

## Checkpoint: predict the transformation

For one mouse:

```text
baseline = 24.1 g
final    = 25.3 g
```

Before calculating:

- What sign should the change have?
- What is its value and unit?
- What does the row represent after adding the column?

::: {.callout-note appearance="simple" icon=false}

- **Sign:** positive.
- **Value:** 1.2 g.
- **Row:** still the same mouse.

Adding a derived variable changes columns, not the observational unit.
:::

::: {.notes}
- Reveal the answer and correct sign or unit errors.
- Transition: retain individual observations when comparing groups.
:::

## Plot individual observations first

```{r}
#| echo: false
#| fig-height: 4.8
#| fig-alt: "Two dot plots with five observations each. Group A spans outcomes 1 to 5, while group B is tightly clustered around 3."

illustrative_outcome_plot +
  geom_jitter(width = 0.08, size = 5, show.legend = FALSE)
```

::: {.notes}
- Neutral illustration: both groups have the same mean but different spread.
- Ask what is visible before the mean: sample size, values, spread and extremes.
:::

## A summary deliberately removes detail

```{r}
#| echo: false
#| fig-height: 4.8
#| fig-alt: "The same two dot plots with diamond mean markers. Both means equal 3 despite the much wider spread in group A."

illustrative_outcome_plot +
  geom_jitter(width = 0.08, size = 5, alpha = 0.45, show.legend = FALSE) +
  stat_summary(
    fun = mean,
    geom = "point",
    shape = 18,
    size = 8,
    show.legend = FALSE
  ) +
  labs(
    caption = "Diamonds show group means; illustration only"
  )
```

Same mean; different evidence.

::: {.notes}
- Ask what is lost if only the diamonds are shown.
- Takeaway: retain observations beside summaries.
:::

## Predict the grouped summary

```r
tibble(
  comparison_group = rep(c("A", "B"), each = 5),
  outcome = c(1, 2, 3, 4, 5, 2.8, 2.9, 3, 3.1, 3.2)
) |>
  group_by(comparison_group) |>
  summarise(
    n = n(),
    mean = mean(outcome),
    .groups = "drop"
  )
```

Before running: how many rows should the result contain?

::: {.notes}
- Ask for the row count and new observational unit before advancing.
- Do not substitute the real treatment comparison.
:::

## Predict the grouped summary

```{r}
#| echo: true

tibble(
  comparison_group = rep(c("A", "B"), each = 5),
  outcome = c(1, 2, 3, 4, 5, 2.8, 2.9, 3, 3.1, 3.2)
) |>
  group_by(comparison_group) |>
  summarise(
    n = n(),
    mean = mean(outcome),
    .groups = "drop"
  )
```

One row now represents one group.

::: {.notes}
- Answer: two rows, one per group.
- Keep `n` beside the mean so the denominator remains visible.
- Source: https://datacarpentry.github.io/R-ecology-lesson/instructor/working-with-data.html#the-split-apply-combine-approach
:::

## Uncertainty belongs with an estimate

| Quantity | What it reports |
|---|---|
| observed difference | size and direction in these data |
| confidence interval | effect sizes compatible with the data and model under the stated procedure |
| p-value | how unusual the result would be if a specified null model were true |

None establishes that the design, data handling or code are correct.

::: {.notes}
- Keep the statistical treatment modest.
- A confidence interval is not the range containing 95% of mice.
- A p-value is neither the probability of the null nor biological importance.
- Do not calculate the exercise's final values.
:::

## Before choosing a test, record the assumptions

- What is the observational unit?
- Are observations independent enough for the proposed comparison?
- How were treatment and measurements assigned or collected?
- Are missing values plausibly ignorable?
- Do spread, shape or unusual values challenge the summary?
- Are all measurements expressed in the documented unit?

::: {.notes}
- Shared cages can challenge a simple independence assumption.
- Cage and batch may contribute variation; do not turn this into a mixed-model lecture.
- Test choice follows the question, design and observed data.
:::

## Visualization is verification evidence

A plot can reveal:

- spread hidden by a mean;
- an observation on the wrong scale;
- a group with few usable values;
- an unexpected relationship between baseline and final weight;
- a transformation whose sign or range is implausible.

The figure should be checked before it is interpreted.

::: {.notes}
- A plot is an analytical result and a diagnostic artifact.
- In the exercise, retain observations and check agreement with numerical results.
:::

## Checkpoint: what does the summary row represent? {.smaller}

After `group_by(treatment)` and `summarise()`:

1. Does one row still represent one mouse?
2. What does `n` establish?
3. Which feature of the individual data can the mean conceal?
4. Which assumption could shared cages challenge?

::: {.callout-note appearance="simple" icon=false style="visibility: hidden;"}

1. One row represents one treatment group—not one mouse.
2. `n` records the denominator used by the summary.
3. A mean can conceal spread, overlap, clusters or outliers.
4. Shared cages can challenge simple independence.
:::

::: {.notes}
- Take answers before advancing.
- Use the response to decide whether the class is ready for the exercise.
:::

## Checkpoint: what does the summary row represent? {.smaller}

After `group_by(treatment)` and `summarise()`:

1. Does one row still represent one mouse?
2. What does `n` establish?
3. Which feature of the individual data can the mean conceal?
4. Which assumption could shared cages challenge?

::: {.callout-note appearance="simple" icon=false}

1. One row represents one treatment group—not one mouse.
2. `n` records the denominator used by the summary.
3. A mean can conceal spread, overlap, clusters or outliers.
4. Shared cages can challenge simple independence.
:::

::: {.notes}
- Reveal each answer and correct the observational-unit language first.
- Takeaway: summaries change representation and remove detail.
:::

## Result is not interpretation

:::: {.columns}

::: {.column width="48%"}
### Result

- observed difference
- uncertainty
- sample size
- visual pattern
:::

::: {.column width="48%"}
### Interpretation

- relation to the biological question
- plausible explanations
- design limits
- remaining uncertainty
:::

::::

::: {.notes}
- Write the numerical and visual result before the biological explanation.
- Avoid causal language beyond the design.
:::

## The design limits the claim

The table records:

- assigned treatment;
- paired weights;
- sex, cage and batch;
- missing measurements.

It does not automatically establish:

- the cause of every individual difference;
- independence within cages;
- why a measurement is missing;
- generality beyond the studied mice and conditions.

::: {.notes}
- Treat limits as properties of the design, not a ritual disclaimer.
- Ask what additional evidence would change one conclusion.
:::

## Why I trust this

| Evidence | What it supports |
|---|---|
| rows match distinct mouse IDs | observational unit represented once |
| missingness recorded by group | effective sample is visible |
| ranges agree with units | scale is plausible |
| derived values inspected | transformation behaved as specified |
| individuals agree with summary | result is not a summary artifact |
| uncertainty and assumptions stated | limits remain visible |

Trust points to evidence. Remaining doubt should also be recorded.

::: {.notes}
- Evidence must be specific and inspectable.
- “I checked carefully” is not evidence.
- Ask for one remaining doubt after every listed check passes.
:::

## AI can explain one step

Give it the local context:

```text
Here is one inspection command and its output.

1. Explain what the command checks.
2. Name one concerning result and why it matters.
3. Do not repair the data or continue the analysis.
```

Then test the explanation against the table or a small code change.

Generated code can reduce typing. The question, context, output and assumptions
still require scientific evaluation.

::: {.notes}
- Keep this bounded to explanation or debugging.
- Retain the command and output beside the AI response; check the explanation with R.
- Use Teams Copilot; do not send sensitive data to an unapproved service.
:::

## The exercise follows the same reasoning sequence

```text
Situation
  → question and prediction
  → how the data arose
  → analysis specification
  → inspect
  → compare final weight
  → derive and compare change
  → verify
  → result
  → interpretation
  → why I trust this
```

::: {.notes}
- This is a map, not a solution.
- The exercise asks about final weight and change so students can distinguish the targets.
- The Quarto document remains the executable record.
:::

## Begin with the question

Open:

```text
exercises/tabular-comparison/analysis.qmd
```

Start with:

1. question and prediction;
2. how the data arose;
3. analysis specification;
4. inspection.

Do not begin by requesting the complete analysis from AI.

::: {.notes}
- If needed: `BIOSCI504::copy_template("lecture-3")` from the course Project.
- Do not overwrite an existing exercise directory.
- Give students time to commit to a prediction before comparison.
:::

## Questions

Then move into the `tabular-comparison` exercise.

::: {.notes}
- Take questions about the reasoning sequence.
- Handle workstation-specific problems individually during the exercise.
:::

# Further reading {.smaller}

- [Data Carpentry: Formatting Data Tables](https://datacarpentry.github.io/spreadsheet-ecology-lesson/01-format-data.html)
- [R Ecology: Exploring and Understanding Data](https://datacarpentry.github.io/R-ecology-lesson/instructor/how-r-thinks-about-data.html)
- [R Ecology: Working With Data](https://datacarpentry.github.io/R-ecology-lesson/instructor/working-with-data.html)
- [R Ecology: Data Visualization With ggplot2](https://datacarpentry.github.io/R-ecology-lesson/instructor/visualizing-ggplot.html)
- [Software Carpentry: Debugging](https://swcarpentry.github.io/python-novice-inflammation/instructor/11-debugging.html)
- [`BIOSCI504` package](https://github.com/cohmathonc/BIOSCI504)

Adapted from Data Carpentry and Software Carpentry instructional materials,
Copyright © The Carpentries, [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).
Changes were made for BIOSCI 504; no endorsement is implied.

::: {.notes}
- Tables and figures use the public `BIOSCI504` package commit recorded in the source.
:::