Field NotesClassroomNo. 23

How to graph plant growth data: line graphs, averages and error bars

Three pots per treatment and a page of heights is enough for an honest graph. Here is how to pick the graph, summarize the plants, show the spread and avoid errors that make a result look stronger than it is.

A hand writes notes and a labeled leaf diagram in a notebook, with potted plants behind

The short answer Graph height against day after planting as a line graph, one line per treatment, and compare final values with a dot or bar graph that shows every plant. Summarize each treatment with its mean, median and range (high schoolers can add standard deviation), and say in the caption what any error bars show and how many plants they cover. Overlapping error bars are not a statistical test.

The graph is where a sheet of heights or germination tallies becomes a claim, and where a careless choice can make a result look stronger than it is. If the class is still collecting, see how to measure plant growth.

Match the graph to the data

The National Center for Education Statistics (NCES) graphing tutorial gives the basic split: line graphs “track changes over short and long periods of time,” bar graphs “compare things between different groups,” and X-Y (scatter) plots “determine relationships between the two different things.”

Your data Graph x-axis y-axis
Same plants measured on several days Line graph: one line per treatment mean, plants as faint dots Day after planting Height (mm)
One final value per plant, by treatment A dot per plant with the mean marked, or bars with dots Treatment Final height (mm) or dry mass (g)
Two things measured on each plant Scatter plot, no connecting line The likely driver, such as light (lux) The response, such as height (mm)
Seeds sprouting over days Line graph of cumulative percent germinated Day after sowing Seeds germinated so far (%)

For germination, plot this each day (plant science fair project ideas has the setup):

cumulative germination (%) = (seeds sprouted so far ÷ seeds sown) × 100

For line and bar graphs, NCES says the x-axis “typically” shows “different time periods or names of things being compared” and the y-axis “the amount of stuff being measured.” That matches the usual convention: independent variable along the bottom, dependent variable up the side. NCES’s check for which is which: “(Independent variable) causes a change in (Dependent Variable).” “Fertilizer causes a change in height” works; the reverse doesn’t.

Set up the sheet, then summarize the plants

Use one row per measurement: day after planting, date, pot ID, treatment, value, unit, notes. Code pots (C1 to C3, F1 to F3), never by student name. Free printable data sheets are on the teacher resources page. Then build a summary table, one row per treatment per day:

  • Mean: “the sum of the data points divided by the number of data points” (NIST).
  • Median: the middle value once sorted; with an even count, the average of the two middle values. NIST notes that extreme values “distort the mean” but “do not distort the median,” which helps when one seedling is stunted.
  • Range: largest minus smallest, “based only on the lowest and highest extreme values” (NIST).
  • n: how many plants are behind the row.

Show the spread, not just the average

A biology paper by Cumming, Fidler and Vaux sorts error bars into two kinds. Descriptive bars show the data: the range, and the standard deviation (SD), the “typical or (roughly speaking) average difference between the data points and their mean.” Inferential bars, the standard error (SE) and confidence interval (CI), show how well the mean is pinned down.

s = √[ Σ(x − x̄)² ÷ (n − 1) ]

where s is the sample standard deviation, x is each plant’s value, x̄ is the treatment mean, Σ means add up over all plants, and n is the number of plants (NIST’s formula).

SE = s ÷ √n

with s and n as above. SD “does not change systematically as n changes”; SE shrinks as you add plants.

The paper’s rules that matter most in class:

  • Say in the caption what the bars are, and give n. Unlabeled bars are “meaningless, or misleading.”
  • Measuring one plant three times is “repetition of measurement on one individual,” so n = 1.
  • With very few plants, “for example n = 3,” it is “better to simply plot the individual data points.”
  • With n = 3, SE bars “must be multiplied by 4” to approximate a 95% CI, so they look tighter than the evidence is.

Worked example: three pots per treatment

These numbers are made up. Final heights on day 14, read to the nearest millimeter:

  • Control: 41, 45, 49 mm
  • Fertilizer: 52, 58, 61 mm

Control. Mean = (41 + 45 + 49) ÷ 3 = 135 ÷ 3 = 45 mm. Median 45 mm, range 49 − 41 = 8 mm. Differences from the mean: −4, 0, +4. Squared: 16, 0, 16, sum 32. Divide by n − 1 = 2: 16. s = √16 = 4.0 mm.

Fertilizer. Mean = (52 + 58 + 61) ÷ 3 = 171 ÷ 3 = 57 mm. Median 58 mm, range 61 − 52 = 9 mm. Differences: −5, +1, +4. Squared: 25, 1, 16, sum 42. Divide by 2: 21. s = √21 ≈ 4.6 mm.

Report 45 ± 4.0 mm and 57 ± 4.6 mm (mean ± SD, n = 3 plants each), and plot all six plants. The dots carry the sturdiest finding: every fertilized plant beat every control, by 3 mm at the closest (52 vs. 49).

For high school, add SE, using the unrounded SDs: 4.0 ÷ √3 ≈ 2.3 mm and 4.58 ÷ √3 ≈ 2.6 mm. Those SE bars (42.7 to 47.3 and 54.4 to 59.6 mm) sit well apart. Multiplied by 4, the approximate 95% CIs (about 36 to 54 and 46 to 68 mm) overlap. Same plants, two different pictures.

What error bars can and can’t tell you

Payton, Greenstone and Schenker warn that researchers “sometimes fall into a trap by trying to use such results as a substitute for a hypothesis test.” Their conclusion: “Whether or not these intervals overlap does not imply the statistical significance of the parameters of interest.” For large samples from two normal populations with the same spread and the same true mean, calling a difference whenever SE bars don’t overlap acts like “a test with a type I error rate of about 15% or 16%,” about three times the usual 5%. In their simulations with 5 or 10 values per group, SE bars overlapped less often, so false alarms were more common still. Requiring 95% intervals to separate is “very conservative.”

In the example, the 95% intervals overlap, yet Cumming’s approximate rule for n = 3 (independent groups only) says that if bars twice the SE long don’t overlap, P < 0.05. They don’t (45 + 2 SE = 49.6 mm; 57 − 2 SE = 51.7 mm). The authors still “don’t recommend using error bars where n = 3”; they give the rules to help readers interpret such figures.

On a growth line graph, each day’s bars come from the same plants, and Cumming’s team calls SE or CI bars “irrelevant to comparisons within the same group.” Use them to compare treatments on one day, not one treatment across days.

Common mistakes to catch before the conclusion

  • Lines across categories. NCES uses lines for change over time. A line from “control” to “fertilizer” to “compost” implies in-between values that don’t exist.
  • A bar axis that doesn’t start at zero. NCES notes the y-axis “usually starts counting at 0.” Start it at 40 mm and the example’s bars stand 5 and 17 mm tall, so fertilized plants look more than three times taller.
  • Missing units. “Height” could be mm, cm or inches. Put the unit in every axis title.
  • Too many decimal places. A spreadsheet can show √21 as 4.5825757; a millimeter ruler doesn’t support that. One decimal place is plenty.
  • Averaging away the spread. A lone mean hides whether one plant drove the result.

Spreadsheet steps that work in any program

  1. Build the summary table with the average, median, minimum, maximum and count functions. If there are two standard deviation functions, use the sample one, which divides by n − 1.
  2. For growth, make a line chart: day after planting as the x values, one series per treatment mean, and, where possible, each plant as a marker-only series.
  3. For final values, plot each plant as a marker above its treatment, with the mean as a bar or short line. A hand-drawn dot plot is just as valid.
  4. For error bars, use custom values from your own SD, SE or minimum and maximum columns, and state the type and n in the caption.

What NGSS expects at each grade band

Grades 6 to 8. NGSS Appendix F asks students to “construct, analyze, and/or interpret graphical displays of data” to “identify linear and nonlinear relationships,” to “apply concepts of statistics and probability (including mean, median, mode, and variability),” and to “distinguish between causal and correlational relationships in data.” The mathematics practice adds “ratio, rate, percent, basic operations, simple algebra.” With plant data: mean, median and range per treatment, percent germination, growth rate in mm per day, and whether the line runs straight or bends.

Grades 9 to 12. The band adds “more detailed statistical analysis,” including “determining function fits to data, slope, intercept, and correlation coefficient for linear fits,” and asks students to weigh “limitations of data analysis (e.g., measurement error, sample selection).” The mathematics practice adds “computational tools for statistical analysis.” Neither grade-band list names standard deviation or error bars; they serve the overview’s line that scientists “calculate the degree of certainty in the results.” Fit a straight line only where a growth curve’s points fall along one, and report its slope in mm per day. For the written claim, how to build a plant science lesson plan has a claim, evidence and reasoning frame.

Sources

  1. APPENDIX F – Science and Engineering Practices in the NGSSNext Generation Science Standards (April 2013 release)
  2. Create a Graph TutorialNCES Kids' Zone, National Center for Education Statistics
  3. 1.3.5.1. Measures of LocationNIST/SEMATECH e-Handbook of Statistical Methods
  4. 1.3.5.6. Measures of ScaleNIST/SEMATECH e-Handbook of Statistical Methods
  5. Error bars in experimental biologyJournal of Cell Biology 177(1):7–11 (Cumming, Fidler and Vaux, 2007), via PubMed Central
  6. Overlapping confidence intervals or standard error intervals: What do they mean in terms of statistical significance?Journal of Insect Science 3:34 (Payton, Greenstone and Schenker, 2003)

General information, not professional advice for a specific plant or crop. Spot something wrong? Email info@terrathink.io.

Keep reading

All field notes

Join the waitlist

New field notes by email, and first word when the hardware goes live.

By joining, you agree to get email updates from Terrathink. Unsubscribe anytime. Privacy policy