Reading Behavior Graphs and Protecting Data Quality

A behavior graph turns a stack of data sheets into one picture. Your team uses that picture to see whether a behavior is changing over time. The picture can only be trusted if two things are true. The numbers placed on it must be accurate. The plan must have been followed closely while those numbers were collected.

In this lesson you will practice four skills. You will place observations correctly on a labeled graph. You will describe a graph's level, trend and variability. You will tell a missing observation from a real zero. You will explain why accurate data does not prove the plan was carried out correctly.

Putting observations on the graph

A line graph has two axes, which are the reference lines you read values from.

The x-axis is the horizontal line along the bottom. It shows time, such as sessions or days.

The y-axis is the vertical line on the left. It shows the amount of behavior.

Label both axes, including the unit. A unit is what each number counts. For example, a y-axis label could read "Hitting (number of times per 30-minute session)." The unit matters because you cannot directly compare counts from sessions of different lengths. A count of 6 in 20 minutes is not the same as 6 in 30 minutes. That is why some graphs show a rate instead, such as times per minute.

Each observation becomes one point. Suppose hitting happened 9 times in Session 1. You would place the point above Session 1 on the x-axis, at 9 on the y-axis. Then you connect each point to the next one with a line.

Missing is not zero. A zero is an observation. It means someone observed for the defined observation period, using the measurement procedure the plan specifies, and the behavior did not occur. In this lesson's examples, the observation period is the whole 30-minute session. A missing observation means no data was collected. If you plot a 0 for a missed session, the graph claims "we watched and saw none." That claim is false, and it creates a drop that never happened.

Do not plot a point for a missed session. Record why it was missed, following your data system and your supervisor's instructions. Graphing systems also differ on one detail: whether the line joins the observed points on either side of the gap, or breaks at the gap. Do not decide this yourself. Follow your data system or ask your supervisor.

Level, trend and variability

When you look at a graph, ask three different questions:

• Level: How high or low do the points sit overall? This is the overall amount of behavior, not how it changed.
• Trend: Which overall direction are the points moving over time: up, down or flat? A trend can contain brief reversals.
• Variability: How much do the points fluctuate from one session to the next?

Example 1: A behavior the plan aims to decrease

The plan aims to reduce hitting. The counts per 30-minute session were 9, 8, 6, 7, 4 and 3.

• Level: The points sit in a middle band overall, between 3 and 9 times per session, about 6 on average.
• Trend: The overall direction is downward. One point rises, from 6 to 7, but the pattern as a whole still falls.
• Variability: The fluctuation is fairly small. Most session-to-session changes are 1 or 2. The largest is 3, from 7 to 4.

A tempting description of level is "the points start high and end low." That sentence describes change over time, which is trend. Level asks how much behavior there is overall.

The plan aims to reduce hitting, so a downward trend is the desired direction.

Now change one fact. Suppose these exact numbers counted independent requests for help, and the plan aims to increase those requests. The line is identical, but the direction is now the opposite of the goal. Up does not automatically mean better. Always check what the y-axis measures and which direction the plan is aiming for.

Example 2: A missed session and one high point

The plan aims to increase requests for a break. Across seven sessions, the counts were 2, 3, 2, not observed, 3, 9 and 3.

Session 4 gets no point. Plotting a zero there would show a false dip.

Session 6 stands out at 9. The next point is back to 3, so the overall level stays near 2 to 3, with one spike. The spike makes this graph more variable: the points fluctuate more than they would without it. It does not create an upward trend. Look at the overall direction: the points before and after the spike sit at about the same height, so the overall direction is roughly flat.

The spike is also worth reporting. It could reflect something unusual that day, or a recording error.

Practice: Name the overall trend

Every session lasted 30 minutes, so the counts can be compared directly. Across six sessions, a client's calling out was counted as 2, 3, 5, 4, 6 and 7 times.

Name the overall trend: upward, downward or flat. Then explain what the change from 5 to 4 tells you.

Answer: Upward, with one brief reversal

The overall trend is upward. The series starts at 2 and 3 and ends at 6 and 7. Four of the five session-to-session changes go up.

The drop from 5 to 4 is a brief reversal. The very next point, 6, is higher than any point before it, so the drop does not change the overall direction. It adds a little variability: the points do not climb perfectly evenly.

A tempting mistake is to let one drop decide the answer and call the series "going down" or "no clear direction." Trend asks about the direction of the whole series. Variability asks how much the points bounce from one session to the next. The small dip answers the second question, not the first.

Naming the trend does not yet tell you whether it is good news. That depends on what the plan aims for, which is a separate question.

Accurate data versus correct implementation

These two ideas answer different questions.

Data accuracy asks: Did we record what actually happened?

Procedural fidelity, also called implementation fidelity, asks: Did we carry out the plan the way it was written?

Imagine the plan says to give a break immediately after each request. Your counts of requests are perfectly correct. However, the break has been coming two minutes late. The graph accurately shows behavior under a procedure that differs from the plan. A flat line might reflect that drift, rather than a plan that fails when it is followed. This is why fidelity problems are addressed before anyone considers changing the plan.

A graph also cannot prove cause. Behavior shifts with illness, tiredness or a change in routine. A downward line shows what changed. It does not show why it changed.

Your role: Record exactly what you observe. Report discrepancies, such as missed sessions, unusual points, or plan steps that were not followed as written. Follow the approved plan and your supervisor's direction. Do not change an intervention because of what a graph seems to show.

Checkpoint

The plan aims to decrease leaving the work area. The counts per 30-minute session for one week were:

• Monday: 5
• Tuesday: 4
• Wednesday: the client was absent
• Thursday: 4
• Friday: 1

A coworker plotted Wednesday as 0. They also say, "Friday dropped, so the plan works; let's skip the extra prompts." You know that on Thursday and Friday a substitute left out one step of the plan.

What is wrong with the graph? What can the Friday point tell you? What should you do?

Answer: Remove the false zero, treat Friday cautiously, and report

Wednesday was never observed, so it should have no point. The false 0 creates a drop that did not happen. Whether the line joins Tuesday to Thursday across the gap depends on your data system or your supervisor's direction.

Friday is a single low point, so on its own it does not establish a trend. It also came from a session where a plan step was skipped. Even if the count was recorded accurately, it does not show that the plan as written produced the change.

You should report the missed session, the substitute's skipped step and the graphing error to your supervisor. Keep following the approved plan, including the prompts. Dropping the prompts would mean changing the intervention based on a graph, and that decision belongs to your supervisor.

Recap

• Label both axes with units, and plot each observation at its session and value.
• Zero means "observed for the defined period, and none occurred." Missing means "not observed," and it gets no point. Follow your data system or supervisor on how the line handles the gap.
• Level is how high or low the points sit overall. Trend is their overall direction, which can include brief reversals. Variability is how much they fluctuate.
• Whether a direction counts as improvement depends on what the y-axis measures and on the plan's goal.
• One unusual point adds to variability but does not establish a trend, and a graph does not prove cause.
• Accurate recording and correct implementation are separate questions. Report problems with either one, and follow your supervisor's direction.

Independent study resource

This study resource is not affiliated with, sponsored by, or endorsed by the Behavior Analyst Certification Board (BACB). BACB, Registered Behavior Technician and RBT are registered trademarks of the Behavior Analyst Certification Board. Source citations do not imply endorsement by their authors or publishers.