Why the Unit Matters
A data set may contain many outcome measurements without containing as many independently treated units. If a teaching method is assigned to classrooms, for example, recording a score for every student does not mean that each student was separately assigned a method. The treatment was allocated at the classroom level; the scores were measured at the student level.
This distinction matters whenever an experiment has more than one relevant unit. The unit that receives a treatment, the unit that can receive a treatment independently, and the unit on which an outcome is recorded may be different. Naming them precisely helps describe what was randomized and what the recorded data represent.
These terms describe roles in a design, not three names for the same thing. A classroom can be the assignment and experimental unit when the whole classroom receives one teaching approach. Individual students in that classroom can be the measurement units if each student takes a test. Conversely, in an experiment that assigns a treatment separately to each person, the person may be all three units.
Terminology can vary somewhat across fields, so state the operational facts: what entity received each treatment assignment, what entity could receive a different assignment independently, and what entity supplied each outcome measurement? Those answers make the design understandable even when a study uses unfamiliar labels.
Find the Units by Following the Treatment and the Outcome
Start with the treatment, not with the rows of a spreadsheet. Ask how the treatment was allocated. If a whole classroom is assigned one teaching method, the classroom is the assignment unit. If each plant plot is allocated a fertilizer condition, the plot is the assignment unit. Then ask what could have been assigned separately: that identifies the experimental unit in the ordinary design sense used here.
Next, follow the outcome. If a score is recorded for each student, the students are the measurement units for that score. If plant height is recorded for each plant, the plants are the measurement units. A measurement unit can be nested within an experimental unit, meaning that several measurement units belong to the same unit that received a treatment. That nesting does not change which entity was assigned the treatment.
A useful way to keep the levels straight is to write a short chain:
- Treatment allocation: which entity gets the condition?
- Independent assignment: which entities can receive conditions independently?
- Outcome recording: which entity has a recorded response?
Do not assume that each line in a data file answers all three questions. A row might describe a student’s score, while the treatment column repeats the classroom’s assigned condition for every student in that classroom. Repeating the label makes the file convenient; it does not create new treatment assignments.
Worked Example: Students Measured Within Classrooms
Hypothetical design. Two classrooms are assigned a new teaching approach, and students in each classroom take the same test. In Classroom A, the scores are 78, 80, and 82 points. In Classroom B, the scores are 70, 74, 78, and 82 points. The example is about identifying units and summaries, not about estimating a treatment effect from only two classrooms.
Assignment and experimental units. The teaching approach is assigned to a whole classroom, so the classrooms are the assignment units. Each classroom receives its condition as a unit; students within the same classroom do not receive independently assigned versions of the teaching approach. Thus, the classrooms are also the experimental units.
Measurement units. The test score is recorded for each student, so students are the measurement units. There are seven student scores, but the design has two classrooms assigned to the approach. The seven scores do not represent seven separate classroom-level assignments.
Check the arithmetic at each level. Classroom A’s mean score is \( (78+80+82)/3=240/3=80 \) points. Classroom B’s mean is \( (70+74+78+82)/4=304/4=76 \) points. The mean of these two classroom means is \( (80+76)/2=78 \) points.
The mean across all seven student scores is instead \( (240+304)/7=544/7=77.714\ldots \), or 77.71 points to two decimal places. These summaries differ because the mean of classroom means gives each classroom equal weight, while the student-level mean gives each student equal weight. Here the classrooms have different numbers of students. Neither calculation changes the assignment unit: both summarize scores from students in classrooms.
Plots, Treatments, and Measured Plants
The same distinction appears in agricultural and laboratory settings. If a fertilizer is spread over an entire plot, plants inside that plot share the plot’s treatment allocation. Measuring each plant separately can provide detailed outcome information, but it does not mean that each plant received an independently assigned plot-level fertilizer condition.
The phrase “smallest entity that can be assigned independently” is important. If fertilizer conditions are assigned plot by plot, two plots could, in principle, receive different conditions. Individual plants within one plot do not independently receive different plot-level conditions simply because each plant has a separate height measurement.
Worked Example: Plants Measured Within Fertilizer Plots
Hypothetical design. Four plots are assigned fertilizer conditions: Plots 1 and 2 receive Fertilizer A, while Plots 3 and 4 receive Fertilizer B. Three plants are measured in each plot. Their heights, in centimetres, are: Plot 1, 18, 20, and 22; Plot 2, 19, 21, and 23; Plot 3, 16, 17, and 18; Plot 4, 20, 21, and 22.
Assignment and experimental units. Fertilizer is allocated to plots, so each of the four plots is an assignment unit. Because plots receive conditions independently, the plots are also the experimental units.
Measurement units. Height is measured for each plant, so the plants are the measurement units. There are \(4\times3=12\) plant-height measurements. The treatments, however, were assigned to four plots, not to 12 separate plants.
Calculate a plot summary. For Plot 1, the mean height is \( (18+20+22)/3=60/3=20 \) cm. For Plot 2 it is \( (19+21+23)/3=63/3=21 \) cm. For Plot 3 it is \( (16+17+18)/3=51/3=17 \) cm, and for Plot 4 it is \( (20+21+22)/3=63/3=21 \) cm.
These four plot means summarize the plant measurements at the level where fertilizer was allocated. The arithmetic does not imply that every design or analysis must use plot means; the point is that a summary can be calculated at one level while the observations were recorded at another. The assignment facts still come from the design.
When the Units Coincide
Not every experiment has separate assignment and measurement units. In an individually randomized experiment, each person may be assigned a treatment separately and then have one outcome recorded. In that case, the person is the assignment unit, experimental unit, and measurement unit. The terms remain useful because they show that the units coincide for a reason, rather than because the labels are interchangeable.
Worked Example: People Assigned and Measured Individually
Hypothetical design. Eight volunteers are assigned individually to one of two conditions, with four people in each condition. Each person has one blood-pressure outcome recorded in mmHg. The outcomes are 118, 122, 120, and 116 for Condition A, and 124, 126, 122, and 120 for Condition B.
Identify the units. Each volunteer receives an individual assignment, so each person is an assignment unit and an experimental unit. Blood pressure is recorded for each person, making each person a measurement unit as well.
Calculate the condition means. For Condition A, \( \bar{x}_A=(118+122+120+116)/4=476/4=119 \) mmHg. For Condition B, \( \bar{x}_B=(124+126+122+120)/4=492/4=123 \) mmHg. These are descriptive means of the recorded outcomes in the two conditions. This small example does not, by itself, establish the size or certainty of a treatment effect.
In this design, four measurements in a condition correspond to four individually assigned people. That correspondence follows from how the volunteers were assigned, not from the fact that the table contains four numbers.
Why the Assignment Level Determines the Independent Units
The reason unit identification matters can be seen by considering what random assignment actually varies. Suppose a treatment is assigned to classrooms. The allocation can change which classrooms receive which conditions. Students within one classroom inherit that classroom’s assignment; they are not separately switched among conditions by the assignment process.
This is a design-based point, not a claim that outcomes from students in the same classroom must be numerically identical. Students can have different scores. The point is that their treatment labels arise from a shared classroom allocation. The same reasoning applies to plants sharing a plot-level treatment.
A good description therefore reports both levels when they differ: for example, “The teaching approach was assigned to classrooms, and test scores were measured for students,” or “Fertilizer was assigned to plots, and height was measured for plants.” Such wording tells the reader what the intervention was allocated to and where the outcomes came from.
Common Mistake: Counting Rows as Independent Assignments
Wrong reasoning: “There are 60 student score rows, so there were 60 independent experimental units.” This conclusion does not follow from the data table. If the treatment was assigned to six classrooms, then the classroom—not each row—is the unit of assignment. The rows describe measurements taken from students within those classrooms.
Correction: Trace the treatment allocation back to the entity that received it. State the assignment and experimental units, then separately state the measurement units. The number of recorded outcomes may exceed the number of independently treated units, and a repeated treatment label in the data does not create additional assignments.
A related mistake is to call a plant the experimental unit merely because it has its own measured height. If fertilizer was allocated to entire plots, the plot is the independently treated unit. Conversely, do not automatically call a whole group the experimental unit when individuals were actually assigned treatments independently. The design, not the setting’s appearance, determines the answer.
Check Your Understanding
For each situation, identify the assignment unit, experimental unit, and measurement unit, and explain your reasoning.
- A school assigns one reading program to each whole classroom. Reading scores are recorded for every student. Which unit receives the program, and which unit supplies the scores?
- A greenhouse assigns watering conditions independently to pots, with one plant in each pot. Leaf area is recorded for each plant. Which units coincide, and why?
- A field experiment allocates fertilizer to plots and records the height of ten plants in each plot. Does the number of plant measurements determine the number of independent treatment assignments? Explain.
- In a hypothetical individually assigned study, each participant receives one of two conditions and has one outcome recorded. Identify the three units and state whether they coincide.
- A spreadsheet repeats the same treatment label on every student row in a classroom. Why does this repetition not show that students were independently assigned?