StatFAS / Blog

ACF plot: What It Shows, When to Use It and How to Interpret the Results

Published November 2026 · Time-series plots for biological, genomic and medical research

Before any time-series model is fitted, the question is how strongly each observation depends on earlier ones. The ACF plot (autocorrelation function plot, also called a correlogram) answers that question lag by lag.

Skip the setup and make the ACF plot now.
Paste the data, get the tables, the ACF plot and APA 7 wording in seconds. Analysis needs no sign-in; downloads need a free sign up, then log in.
Open the free ACF plot calculator

What is a ACF plot?

An ACF plot shows the correlation between a series and lagged copies of itself. The x-axis is the lag (1 step back, 2 steps back and so on) and the y-axis is the autocorrelation coefficient r, which ranges from −1 to +1. Each vertical stem is one lag. Dashed horizontal lines mark approximate 95% limits, ±1.96/√N, the range expected when the series is pure white noise.

At lag k the coefficient is rk = Σ(xt − x̄)(xt+k − x̄) / Σ(xt − x̄)². A Ljung–Box test adds one overall verdict by testing many lags together.

When and why to use a ACF plot

When

Why

Typical biomedical examples

daily ICU admissions, continuous glucose readings, heart rate variability series, daily case counts, repeated clinic measurements over time.

How to interpret the results

Slow, steady decay from a high first value

This pattern signals a trend or a nonstationary mean. Difference the series once and draw the ACF plot again.

Spikes at seasonal lags

A stem at lag 12 (and 24, 36) in monthly data signals yearly seasonality. A seasonal difference at that lag is the usual remedy.

Cutoff after lag q

When stems are significant up to lag q and then fall inside the limits, a moving-average component of order q is suggested, especially when the PACF tails off.

Smooth or wavy decay

Geometric or sinusoidal decay suggests an autoregressive component. Consult the PACF plot for the order.

All stems inside the limits

The series is consistent with white noise. About 1 lag in 20 is expected to cross the limits by chance, so one stem barely beyond the line at a high lag is not meaningful.

Limits are pointwise

Each dashed limit applies to one lag alone. The Ljung–Box test judges a block of lags together and is better for an overall decision. When the data are autocorrelated, Bartlett’s wider limits are more accurate than ±1.96/√N.

Strongly negative first value after differencing

A lag-1 coefficient near −0.5 after differencing can mean the series was over-differenced.

Linear dependence only

The ACF plot measures linear association between time points. A significant stem is not evidence of causation.

Common mistakes

How to make a ACF plot

The easy way: StatFAS.com

The free ACF plot calculator at StatFAS.com handles the calculations, draws the figure to publication standard and writes the APA 7 results text. The ACF plot calculator page walks through every step.

Make the ACF plot in your browser.
Paste the data, get the tables, the ACF plot and APA 7 wording in seconds. Analysis needs no sign-in; downloads need a free sign up, then log in.
Open the free ACF plot calculator

SPSS

  1. If seasonal differencing is wanted, define the cycle first with Data > Define Date and Time.
  2. Choose Analyze > Forecasting > Autocorrelations.
  3. Move the series into Variables.
  4. Under Display tick Autocorrelations (tick Partial autocorrelations as well to get the PACF plot in the same run).
  5. Under Transform choose Difference (1) or Seasonally difference when needed.
  6. Click Options, set the maximum number of lags, and choose the standard error method (Independence model or Bartlett’s approximation).
  7. SPSS draws the plot and prints a table with the Box–Ljung statistic for every lag.

R

x <- ts(values, frequency = 12)
acf(x, lag.max = 24, main = "")            # lag 0 (always 1) is drawn first
Box.test(x, lag = 24, type = "Ljung-Box")  # joint test over 24 lags

acf(diff(x), lag.max = 24)                 # after first difference
acf(diff(x, lag = 12), lag.max = 24)       # after seasonal difference

Python

import matplotlib.pyplot as plt
from statsmodels.graphics.tsaplots import plot_acf
from statsmodels.stats.diagnostic import acorr_ljungbox

plot_acf(y, lags=24, zero=False, bartlett_confint=False, alpha=0.05)  # white-noise limits
plt.show()
print(acorr_ljungbox(y, lags=[24]))

plot_acf(y.diff().dropna(), lags=24, zero=False)      # first difference
plot_acf(y.diff(12).dropna(), lags=24, zero=False)    # seasonal difference
plt.show()

How to report the results in APA 7

Example template: “The ACF of the [variable] series (N = [N]) was computed for [K] lags with 95% limits of ±[limit]. Significant autocorrelation occurred at lags [list]. A Ljung–Box test indicated [significant / no significant] serial dependence, Q([K]) = [value], p [value].” Mention any differencing that was applied before the ACF plot was drawn.

Frequently asked questions

What does the dashed line on an ACF plot mean?

The dashed lines mark approximate 95% limits for white noise, ±1.96/√N. A stem outside the lines is larger than expected for an uncorrelated series.

How many lags should be shown in an ACF plot?

For non-seasonal data 20 to 24 lags are usual, and for seasonal data at least two full cycles (24 for monthly data). Do not exceed about half of the series length.

What is the difference between the ACF plot and the PACF plot?

The ACF plot shows the total correlation at each lag. The PACF plot shows the correlation at each lag after the effect of shorter lags is removed. Together the two plots suggest the model order.

Does a significant ACF stem mean the series is not stationary?

Not by itself. Stationary series can have significant stems. A slow decay over many lags is the sign of nonstationarity.

Key takeaways

Related reading

PACF plot blog · Seasonal decomposition plot blog · ACF plot calculator