Skip to contents

Sequential t-Test

Meike Snijder-Steinhilber

2026-01-27

Source: vignettes/t_test.Rmd
t_test.Rmd

What is the sequential t-test?

The sequential t-test is based on the Sequential Probability Ratio Test (SPRT) by Abraham Abraham Wald (1947), which is a highly efficient sequential hypothesis test. However, the usage of Wald´s SPRT is limited in the case of normally distributed data, because the variance has to be known or specified in the hypothesis. Rushton (1950; 1952) and Hajnal (1961) have further developed the SPRT using the t-statistic. The basic idea is to transform the sequence of observations (which is dependent on the variance) into a sequence of the associated t-statistic (which is independent of the variance).

In the SPRT the null and alternative hypotheses are defined as follows, with \(\theta\) representing the model parameter :

\[ H_0:\ \theta\ =\ \theta_0, \]

\[ H_1:\ \theta\ =\ \theta_1. \]

The test statistic of the SPRT is based on a likelihood ratio, which is a measure of the relative evidence in the data for the given hypotheses. More specifically, it is the ratio of the likelihood of the alternative hypothesis to the likelihood of the null hypothesis at the m-th step of the sampling process (LRm).

\[ LR_{m} = \frac{f(\text{data}_m | H_1)}{f(\text{data}_m | H_0)} = \frac{f(x_1,...,x_m | \theta_1)}{f(x_1,...,x_m | \theta_0)} \]

Before the transformation into the t-statistic, the model parameter \(\theta\) contains the parameters of a normal distribution: the mean (\(\mu\)) and the standard deviation (\(\sigma\)). Therefore, the Wald SPRT requires prior knowledge about the variance (\(\sigma^2\)) or a specification in the hypotheses.

After the transformation of the observed values into the associated t-statistic, the model parameter \(\theta\) contains the parameters of the non-central t-distribution: the degrees of freedom (df) and the non-centrality parameter (\(\Delta\)).

\[ {f(x_1,...,x_m | \mu,\sigma)} => {f(t_2,...,t_m | df,\Delta)} \]

For the calculation of the degrees of freedom, only the sample size of the group(s) is needed. The non-centrality parameter also requires a specification of the expected effect size in form of Cohen’s d.

To eventually calculate the LR of the sequential t-test, only the current tm-statistic is necessary. S. Rushton (1950) demonstrated that an SPRT can be performed by simply considering the ratio of probability densities for the most recent tm statistic under the alternative and null hypothesis at any m-th stage. Thus, the test statistic for a one and two-sided sequential t-test can be calculated as follows:

\[ LR_{m,\ one-sided\ sequential\ t-test} = \frac{f(t_m | \theta_1)} {f(t_m | \theta_0)}, \]

\[ LR_{m,\ two-sided\ sequential\ t-test} = \frac{f(t_m^2 | \theta_1)} {f(t_m^2 | \theta_0)}. \]

To account for the fact that the algebraic sign is unknown in a two-sided test, the t-value is squared (S. Rushton, 1952).

After the calculation of the test statistic, the decision will be either to continue sampling or to terminate the sampling and accept one of the hypotheses. A. Wald (1945) defined the following rules for the SPRT:

Condition Decision
LRm ≤ B accept H0 and reject H1
B < LRm < A continue sampling
LRm ≤ A accept H1 and reject H0

The A and B boundaries are calculated with the previously defined error rates \(\alpha\) (Type I error) and \(\beta\) (Type II error) as follows:

\[ A = \left( \frac{1 - \beta}{\alpha} \right), \]

\[ B = \left( \frac{\beta}{1 - \alpha} \right). \]

In summary, three specifications are required to calculate a sequential t-test:

  • the \(\alpha\) error probability (usually 0.05 or less),

  • the \(\beta\) error probability (usually .20 or less), and

  • Cohen´s d (either as the expected effect size or as the lower limit for a substantial effect).

How do I use the seq_ttest() function?

The seq_ttest() function has arguments to specify the requested sequential t-test. The table below shows all possible combinations which can be performed with the package.

Two-sample test One-sample test
Two-sided x x
One-sided x x
Paired (repeated measures) x

Other recommended vignettes cover:

Argument

The seq_ttest() function works similarly to the t.test() function from the stats package if one is familiar with that already.

Sequential t-tests require some specification from the user:

  • the variables, which contain the data,

  • the error probability alpha,

  • the power (1 - \(\beta\)),

  • the effect size Cohen’s d, which represents the expected effect size or the lowest effect size of interest, and

  • optional arguments to further specify the test.

However, in some cases, it is not necessary to specify all arguments because some of them have default values. If these values are the ones required, they can be skipped.

Argument Default value Input option
x - formula or numeric input
y NULL numeric vector
data NULL data frame
mu 0 numeric value
d - numeric value
alpha 0.05 numeric value between 0 and 1
power 0.95 numeric value between 0 and 1
alternative “two.sided” “two.sided”, “greater”, “less”
paired FALSE TRUE or FALSE
na.rm TRUE TRUE or FALSE
verbose TRUE TRUE or FALSE

There are two different ways how the data can be transferred into the function. The x argument takes either formula or numeric input. Which input option is recommended depends on the structure of the data.

x argument: formula input

The formula input is used when both groups are merged in one variable and there is a second variable that indicates group membership. This input option uses the x argument and the data argument if the variables are stored in a data frame.

Two-sample test 

library(sprtt)
## Note: Sample size planning functions require simulation data (~70 MB).
## Data will download automatically on first use.
## See ?download_sample_size_data for more information.
# show data frame --------------------------------------------------------------
head(df_income)
##   monthly_income    sex
## 1       4091.001 female
## 2       3274.591   male
## 3       2696.436 female
## 4       3826.413   male
## 5       3522.478 female
## 6       2563.597   male

# sequential t-test: data argument ---------------------------------------------
seq_ttest(monthly_income ~ sex,     # x argument 
          data = df_income,
          d = 0.2)
## 
## *****  Sequential  Two Sample t-test *****
## 
## formula: monthly_income ~ sex
## test statistic:
##  log-likelihood ratio = -0.594, decision = continue sampling
## SPRT thresholds:
##  lower log(B) = -2.944, upper log(A) = 2.944
## Log-Likelihood of the:
##  alternative hypothesis = 0.827
##  null hypothesis = 1.421
## alternative hypothesis: true difference in means is not equal to 0.
## specified effect size: Cohen's d = 0.2
## degrees of freedom: df = 118
## sample estimates:
## mean of x mean of y 
##  3072.086  3080.715 
## *Note: to get access to the object of the results use the @ or [] instead of the $ operator.

One-sample test

To perform a one-sample test, the right side of the formula has to be 1. The mu argument is also required, which specifies the mean value that one wants to test against.

# show data frame --------------------------------------------------------------
head(df_income)
##   monthly_income    sex
## 1       4091.001 female
## 2       3274.591   male
## 3       2696.436 female
## 4       3826.413   male
## 5       3522.478 female
## 6       2563.597   male

# sequential t-test: data argument ---------------------------------------------
seq_ttest(monthly_income ~ 1,     # x argument 
          mu = 3000,
          d = 0.5,
          data = df_income)
## 
## *****  Sequential One Sample t-test *****
## 
## formula: monthly_income ~ 1
## test statistic:
##  log-likelihood ratio = -6.288, decision = accept H0
## SPRT thresholds:
##  lower log(B) = -2.944, upper log(A) = 2.944
## Log-Likelihood of the:
##  alternative hypothesis = -9.182
##  null hypothesis = -2.894
## alternative hypothesis: true mean is not equal to 3000.
## specified effect size: Cohen's d = 0.5
## degrees of freedom: df = 119
## sample estimates:
## mean of x 
##    3076.4 
## *Note: to get access to the object of the results use the @ or [] instead of the $ operator.

x argument: numeric input

The numeric input is used when each group has its own variable. The variables can either be put in the global environment directly or be stored in a data frame.

Two-sample test

If one wants to perform a two-sample test, the y argument is required in addition to x. If the data are stored in a data frame, the $ operator is essential to get access to the variables.

# show data frame --------------------------------------------------------------
head(df_cancer)
##   treatment_group control_group
## 1        6.097665      4.493064
## 2        6.609234      5.520956
## 3        5.665810      3.954091
## 4        4.830564      3.733212
## 5        4.917361      4.109373
## 6        3.457433      3.563800

# sequential t-test: $ operator ------------------------------------------------
seq_ttest(df_cancer$treatment_group, # x argument
          df_cancer$control_group,   # y argument
          d = 0.3,                   
          verbose = FALSE)           
## 
## *****  Sequential  Two Sample t-test *****
## 
## formula: df_cancer$treatment_group and  df_cancer$control_group
## test statistic:
##  log-likelihood ratio = 10.777, decision = accept H1
## SPRT thresholds:
##  lower log(B) = -2.944, upper log(A) = 2.944


# sequential t-test: global variables ------------------------------------------
treatment <- df_cancer$treatment_group
control <- df_cancer$control_group

seq_ttest(treatment,
          control,
          d = 0.3,
          verbose = FALSE)
## 
## *****  Sequential  Two Sample t-test *****
## 
## formula: treatment and  control
## test statistic:
##  log-likelihood ratio = 10.777, decision = accept H1
## SPRT thresholds:
##  lower log(B) = -2.944, upper log(A) = 2.944

One-sample test

If one wants to perform a one-sample test there is only one group and therefore only one variable. If the data are in a data frame, the $ operator is essential to get access to the variables. The mu argument is additionally required, which specifies the mean which one wants to test against.

# show data frame --------------------------------------------------------------
head(df_cancer)
##   treatment_group control_group
## 1        6.097665      4.493064
## 2        6.609234      5.520956
## 3        5.665810      3.954091
## 4        4.830564      3.733212
## 5        4.917361      4.109373
## 6        3.457433      3.563800

# sequential t-test: $ operator ------------------------------------------------
seq_ttest(df_cancer$treatment_group,     # x argument
          mu = 3.5,            
          d = 0.2,
          verbose = FALSE)
## 
## *****  Sequential One Sample t-test *****
## 
## formula: df_cancer$treatment_group
## test statistic:
##  log-likelihood ratio = 16.677, decision = accept H1
## SPRT thresholds:
##  lower log(B) = -2.944, upper log(A) = 2.944


# sequential t-test: global variables ------------------------------------------
treatment <- df_cancer$treatment_group

seq_ttest(treatment,     # x argument
          mu = 3.5, 
          data = df,
          d = 0.2,                  
          verbose = FALSE)
## 
## *****  Sequential One Sample t-test *****
## 
## formula: treatment
## test statistic:
##  log-likelihood ratio = 16.677, decision = accept H1
## SPRT thresholds:
##  lower log(B) = -2.944, upper log(A) = 2.944

Further arguments

Paired

The paired argument states if the data are paired. To perform a paired sequential t-test, paired has to be set to TRUE.

# show data frame --------------------------------------------------------------
head(df_stress)
##   baseline_stress one_year_stress
## 1        7.175250        7.844337
## 2        4.918343        5.527191
## 3        4.634266        5.783046
## 4        5.671340        7.793956
## 5        4.141257        3.133889
## 6        4.771696        8.548586

# sequential t-test: $ operator ------------------------------------------------
seq_ttest(df_stress$baseline_stress, # x argument
          df_stress$one_year_stress, # y argument
          d = 0.3,
          paired = TRUE,
          data = df_stress)  
## 
## *****  Sequential Paired t-test *****
## 
## formula: df_stress$baseline_stress and  df_stress$one_year_stress
## test statistic:
##  log-likelihood ratio = 7.174, decision = accept H1
## SPRT thresholds:
##  lower log(B) = -2.944, upper log(A) = 2.944
## Log-Likelihood of the:
##  alternative hypothesis = -3.483
##  null hypothesis = -10.657
## alternative hypothesis: true difference in means is not equal to 0.
## specified effect size: Cohen's d = 0.3
## degrees of freedom: df = 119
## sample estimates:
## mean difference 
##        -0.39622 
## *Note: to get access to the object of the results use the @ or [] instead of the $ operator.

Alternative

The alternative argument states in which way the test is to be performed:

  • two-sided: "two.sided" or

  • one-sided: "less" or "greater".

# show data frame --------------------------------------------------------------
head(df_income)
##   monthly_income    sex
## 1       4091.001 female
## 2       3274.591   male
## 3       2696.436 female
## 4       3826.413   male
## 5       3522.478 female
## 6       2563.597   male

# sequential t-test: data argument ---------------------------------------------
seq_ttest(monthly_income ~ 1,     # x argument 
          mu = 1000,
          d = 0.3,
          alternative = "two.sided",
          data = df_income)
## 
## *****  Sequential One Sample t-test *****
## 
## formula: monthly_income ~ 1
## test statistic:
##  log-likelihood ratio = 31.548, decision = accept H1
## SPRT thresholds:
##  lower log(B) = -2.944, upper log(A) = 2.944
## Log-Likelihood of the:
##  alternative hypothesis = -149.747
##  null hypothesis = -181.295
## alternative hypothesis: true mean is not equal to 1000.
## specified effect size: Cohen's d = 0.3
## degrees of freedom: df = 119
## sample estimates:
## mean of x 
##    3076.4 
## *Note: to get access to the object of the results use the @ or [] instead of the $ operator.

Na.rm

The na.rm argument defines the handling of missing values. If set to TRUE (default value), it will remove all missing values automatically. If this behavior is not wanted, the na.rm argument has to be set to FALSE. If missing values are discovered, an error is triggered.

Verbose

The verbose argument defines the extent of the output shown in the console. If set to TRUE (default value), the output will be elaborate, if set to FALSE the output will be short.

# sequential t-test: verbose FALSE ---------------------------------------------
seq_ttest(df_cancer$treatment_group, # x argument
          df_cancer$control_group,   # y argument
          d = 0.3,
          verbose = FALSE)
## 
## *****  Sequential  Two Sample t-test *****
## 
## formula: df_cancer$treatment_group and  df_cancer$control_group
## test statistic:
##  log-likelihood ratio = 10.777, decision = accept H1
## SPRT thresholds:
##  lower log(B) = -2.944, upper log(A) = 2.944

# sequential t-test: verbose TRUE ----------------------------------------------
seq_ttest(df_cancer$treatment_group, # x argument
          df_cancer$control_group,   # y argument
          d = 0.3,
          verbose = TRUE) 
## 
## *****  Sequential  Two Sample t-test *****
## 
## formula: df_cancer$treatment_group and  df_cancer$control_group
## test statistic:
##  log-likelihood ratio = 10.777, decision = accept H1
## SPRT thresholds:
##  lower log(B) = -2.944, upper log(A) = 2.944
## Log-Likelihood of the:
##  alternative hypothesis = -11.635
##  null hypothesis = -22.411
## alternative hypothesis: true difference in means is not equal to 0.
## specified effect size: Cohen's d = 0.3
## degrees of freedom: df = 238
## sample estimates:
## mean of x mean of y 
##   4.92417   4.02215 
## *Note: to get access to the object of the results use the @ or [] instead of the $ operator.

How do I get access to the results?

The simplest way to get access to the results is to run the seq_ttest() function. It will print the results automatically in the console. The verbose argument specifies how much information is wished to be shown.

However, the recommended way is to save the results in an object (e.g “results”). Doing so allows running further calculations with it afterward. It is important to keep in mind that the output object will be an S4 class. Therefore the access operator is the @ sign or the [] brackets.

# save the resuts in a object 
results <- seq_ttest(df_cancer$treatment_group,
                     df_cancer$control_group,  
                     d = 0.3)
# access the object with the @ operator 
results@decision
## [1] "accept H1"

# access the object with the [] brackets
results["decision"]
## [1] "accept H1"

References

Hajnal, J. (1961). A two-sample sequential t-test. Biometrika, 48(1/2), 65–75. https://doi.org/10.2307/2333131
Rushton, S. (1950). On a sequential t-test. Biometrika, 37, 326–333. https://doi.org/10.2307/2332385
Rushton, S. (1952). On a Two-Sided Sequential t-Test. Biometrika, 39(3/4), 302. https://doi.org/10.2307/2334026
Wald, A. (1945). Sequential tests of statistical hypotheses. The Annals of Mathematical Statistics, 16(2), 117–186.
Wald, Abraham. (1947). Sequential analysis. Wiley.