What Is a Confidence Interval? How to Read and Interpret It in SPSS

What Is a Confidence Interval

You open your SPSS output and see two columns labelled Lower Bound and Upper Bound, or a result reported as 95% CI [0.184, 0.527]. These two numbers form a confidence interval, usually abbreviated as CI. A confidence interval describes a plausible range of values for a population parameter based on the sample data you collected.

For example, your sample has a mean satisfaction score of 3.82 and a 95% confidence interval from 3.70 to 3.94. Here, 3.70 is the lower bound and 3.94 is the upper bound. You can report that the observed mean is 3.82, with a 95% confidence interval from 3.70 to 3.94. This interval does not replace the sample mean, and it does not mean that 95% of respondents fall within that range.

For a simple estimate, the formula usually has the following form:

Point estimate ± critical value × standard error

The point estimate may be a mean, regression coefficient, or indirect effect. The critical value depends on the confidence level and the distribution used. The standard error reflects how much the estimate would vary if you repeatedly drew samples. The significance level and confidence interval are directly related. For example, a 95% confidence level commonly corresponds to a 5% significance level. You can also read What Is a Significance Level if you are mixing up these two concepts.

Why Confidence Intervals Matter in Quantitative Research

A confidence interval tells you how stable a result is. A narrow interval indicates a smaller standard error for the estimate in the sample and model under consideration. A wide interval indicates more uncertainty, which is often related to a small sample size, dispersed data, or an unstable measurement.

In regression, a confidence interval helps you assess the plausible range of an effect coefficient. If coefficient B is 0.36 and the 95% CI ranges from 0.12 to 0.58, the estimated effect is positive and both bounds are greater than 0. If the 95% CI ranges from -0.04 to 0.51, the interval contains 0. You should then be cautious about concluding that the independent variable has an effect different from 0.

In mediation analysis, the confidence interval for the indirect effect is usually assessed with bootstrapping. If the bootstrap confidence interval does not contain 0, the indirect effect is considered significant under this method's rule (Preacher and Hayes, 2008). This is why, when you run PROCESS in SPSS, you need to look at the Indirect effect(s) of X on Y table rather than only looking at the p-value for each regression equation.

A confidence interval also makes your results more transparent. A p-value answers a question about evidence against the null hypothesis. A CI shows both the direction of the effect and the uncertainty around its size. When writing Chapter 4, report the coefficient, confidence interval, and p-value in the same paragraph when the software provides all three statistics.

A CI result should not be separated from the research design. A confidence interval from survey data does not prove that the result is certainly true for every target group. It reflects the uncertainty of the estimate under the model, sampling method, and assumptions you used.

What Confidence Interval Is Acceptable

The question “what confidence interval is good” has no single number that applies to every variable. You need to read a confidence interval according to three factors: the selected confidence level, the parameter being estimated, and whether the interval contains the value being tested. A 95% CI is common in theses, but using 95% does not automatically make a model acceptable.

The table below summarises common rules. Each rule applies only to the situation shown and should not be transferred mechanically to every type of test.

The practical answer to “what confidence interval is acceptable” is therefore to check whether the lower and upper bounds include the value being tested. For an effect coefficient, that value is usually 0. For an odds ratio or hazard ratio, the reference value is usually 1. These two types of parameters must not be read against the same reference point.

A narrow interval is not necessarily better if the model is misspecified, the scale has problems, or the sample is unrepresentative. Conversely, a wide interval does not mean that the data are useless. It indicates that you need to report the uncertainty and review the sample size, dispersion, and model specification instead of simply trying to make the interval narrower.

How to Read a Confidence Interval in the Output

First, identify the type of table you are reading. In SPSS, regression results are usually shown in the Coefficients table, while the CI may appear after you select the relevant statistics option. For mediation analysis with PROCESS, the bootstrap CI appears in the indirect-effects table. For some mean tests, the CI may appear in the One-Sample Test or Independent Samples Test table.

Suppose you run a regression with dependent variable Y and independent variable X. The table below is illustrative output, not the result of an actual study.

For the X row, coefficient B is 0.364. The lower bound is 0.183 and the upper bound is 0.545. Both bounds are positive, so the interval does not contain 0. If the model assumptions are appropriate, you can interpret this as a positive relationship between X and Y in this model.

The Sig. column displays 0.000 in SPSS, but you should not write p = 0.000. The appropriate format is p < 0.001, because the p-value is not actually equal to 0. CI and p-value usually lead to the same decision when they are constructed under the same assumptions and at the same testing level. You still need to check that you are reading the correct table and the correct type of CI.

If the output is based on bootstrapping, pay attention to the column names. You may see BootLLCI and BootULCI, which are the lower and upper bounds of the bootstrap confidence interval. For example, the indirect effect is 0.112, BootLLCI is 0.041, and BootULCI is 0.205. Both bounds are positive and the interval does not contain 0, so the indirect effect is significant under the bootstrap criterion (Preacher and Hayes, 2008).

In SmartPLS, the Path Coefficients table usually contains the columns Original Sample, Sample Mean, Standard Deviation, T Statistics, and P Values. After you run Bootstrapping, the results may provide the lower and upper bounds of the confidence interval. In PLS-SEM, report the bootstrap procedure and number of subsamples clearly. The procedure using 5,000 bootstrap subsamples is described in (Hair et al., 2022).

You can compare your reading of the CI with related concepts such as What Is 95 Reliability, What Is df, and What Are Degrees of Freedom. df can affect the critical value in some formulas, but df is not a confidence interval.

What to Do When the Confidence Interval Is Not Acceptable

First, define what “not acceptable” means in your analysis. If the CI for a coefficient contains 0, this usually means that there is not enough evidence to conclude that the coefficient differs from 0 at the selected significance level. This is not a software error and is not a valid reason to delete a variable simply because the result does not support the hypothesis.

Check How You Read the Output

Check the correct variable row, type of confidence interval, and confidence level. In regression, do not use the CI for Constant to draw a conclusion about the independent variable. In PROCESS, do not use the CI for the total effect as a substitute for the indirect effect. In SmartPLS, make sure you are viewing the results after Bootstrapping, not only the initial estimate.

Check the Data and Variable Coding

Open the .sav or .csv file again and check missing values, reverse-coded items, data types, out-of-range values, and invalid response rows. An item that has not been reverse-coded can reduce scale quality, which can then affect the coefficient and confidence interval. Record every change. Do not edit the file directly and run the analysis again without saving the previous version.

Check the Assumptions and Model

For regression, review linearity, residuals, outliers, multicollinearity, and error independence. A Durbin-Watson statistic between 1 and 3 is a commonly used reference rule when checking error independence (Field, 2013). Read VIF together with the correlations among variables and the model structure. Do not look at one number alone and draw a conclusion.

Report the Result as It Appears in the Data

If the CI still contains 0 after reasonable checks, keep that result. You can write that the 95% confidence interval for the coefficient of X ranges from the lower bound to the upper bound and includes 0. Therefore, the hypothesis that the effect differs from 0 is not supported. You may remove a variable, change the model, or increase the sample size only when there is a clear methodological reason. Do not do so merely to make the CI look better.

Distinguishing Confidence Intervals from p-Values and Standard Deviation

Confidence intervals and p-values are related, but they provide different information. A p-value focuses on this question: if the null hypothesis is true under the model, how unusual are the observed data? A CI shows the range of parameter values that are compatible with the data and the selected confidence level.

The standard deviation describes how dispersed observations are around the mean. The standard error describes how much an estimate, such as a sample mean, would vary across repeated samples. A confidence interval is usually calculated from the standard error, not taken directly from the standard deviation of each response.

A common mistake is to say that “95% of the data lie within the 95% confidence interval.” This is incorrect in the usual meaning of a CI. A confidence interval is for the parameter being estimated. An interval containing most of the data is a different concept, usually related to dispersion or a prediction interval.

You also need to distinguish a CI from scale reliability. Cronbach's Alpha evaluates the internal consistency of items for the purpose of scale analysis, while a CI describes the uncertainty of an estimate. A high Alpha does not automatically make the confidence interval for a regression coefficient narrower.

Common Errors

Reporting the confidence interval as a single number. A CI must have a lower bound and an upper bound. If you write only CI = 95%, you have stated the confidence level but not the result interval.

Concluding that there is a 95% probability that the parameter lies within the interval. Under the frequentist interpretation, the interval is constructed from a long-run sampling procedure. In your thesis, write that the 95% confidence interval for the parameter is estimated from the lower bound to the upper bound. Avoid assigning a direct probability to a fixed parameter.

Seeing that the CI contains 0 and concluding that the model has no value. A statistically nonsignificant coefficient does not mean that the entire model has failed. You need to consider the research objective, other paths, effect size, and data limitations.

Changing 95% to 90% to make the result significant. The confidence level should be decided according to the design and explained in advance or transparently in the report. Do not select the level after seeing the result simply to remove 0 from the interval.

Mixing up a regular CI and a bootstrap CI. The two intervals may be based on different estimation procedures. State the method, the number of bootstrap repetitions when applicable, and the output table used.

Confusing the lower and upper bounds. For a negative effect, such as an interval from -0.48 to -0.09, both bounds are negative and the interval still does not contain 0. Do not see a negative number and conclude that the effect is not significant.

Frequently asked questions

What is a confidence interval in SPSS?

A confidence interval in SPSS is an interval with a lower bound and an upper bound that describes a plausible range for an estimate, such as a mean or regression coefficient. Read the table name, column name, and specific variable row instead of looking at one number in isolation.

What confidence interval is good?

There is no fixed width that applies to every study. For a coefficient tested against 0, you usually check whether the interval contains 0. A narrower interval generally indicates a more precise estimate in that sample, but you must read it together with the design, sample size, and model assumptions.

Is a 95% confidence interval the same as a 0.05 significance level?

In many two-tailed tests, a 95% CI corresponds to a 0.05 significance level. You still need to check the calculation method, whether the test is one-tailed or two-tailed, and which type of CI is being reported. You can also read What Is 95 Reliability to distinguish the confidence level from the precision of the estimate.

If the CI contains 0, can I accept the H1 hypothesis?

If the CI for a coefficient contains 0, you usually do not have enough evidence to conclude that the coefficient differs from 0 at the corresponding significance level. In your thesis, report the interval, p-value, and hypothesis context accurately instead of automatically changing the data or deleting the variable.

Why does the bootstrap confidence interval not contain 0 when the p-value is greater than 0.05?

The two results can differ because of the type of interval, bootstrap procedure, one-tailed test, or rounding. Check the correct table, confidence level, number of bootstrap subsamples, and inferential rule you selected. For an indirect effect, follow the bootstrap CI rule stated in the method and report the result consistently.

Can a confidence interval replace a p-value?

A confidence interval can provide richer information because it shows the direction and range of an effect, but it does not always fully replace the p-value when your university or supervisor requires both. The safest approach is to report the coefficient, CI, and p-value when the output provides these statistics.

Open the Coefficients table or the bootstrap table in your file, record the lower bound, upper bound, and reference value such as 0 separately, then write one interpretation sentence before moving to the next results table. If you need to run and interpret the analysis on your own .sav or .csv file, you can use DoThesis M4 analysis for this step.