The posttest-only control group design is a basic experimental design where participants get randomly assigned to either receive an intervention or not, and then the outcome of interest is measured only once after the intervention takes place in order to determine its effect.
The intervention can be:
- a medical treatment
- a training program
- an exposure to a risk factor, etc.
Note that this design differs from the pretest-posttest randomized controlled trial by having no measurements taken before the intervention. Also, if we remove the random assignment component from this design (and let participants get assigned to groups according to their choosing or that of the researcher), we get the static-group comparison design which is a type of quasi-experiment.
Advantages of the posttest-only control group design
1. The treatment and control groups are equivalent at baseline
This is very important because when the 2 groups are equal, any difference in the outcome measured will be attributed only to the intervention and not to the initial difference between the groups.
Note that these groups are not expected to be perfectly equal, in fact they don’t need to be. As long as we are assigning participants at random, we won’t be subject to selection bias and we will obtain comparable groups.
This is an advantage over static-group comparison where assignment is not done at random, so the initial difference in the characteristics of both groups can be an important factor influencing the outcome of the study and therefore an important source of bias.
2. External factors are controlled
- The use of a control group controls history (i.e. adjusts for the effects of events that can happen at the same time as the intervention and can influence the outcome, therefore becoming a rival hypothesis and a potential source of bias).
- The use of 1 measurement only controls factors related to the instruments used to measure the outcome (since the device is only used once to measure the outcome, we won’t have to deal with changes that can happen to the device or to the quality of measurements from 1 measurement to the next).
- The simultaneity in measuring both groups controls factors that change with time (as these may also affect the outcome and bias the study).
3. Can be used when participants’ anonymity must be kept
When subjects are measured before and after the intervention, some sort of system should be installed in order to know which measurement corresponds to which participant. In these designs, the participant’s ID, name or phone number will be recorded in a database which may interfere with some type of studies where the participant’s anonymity is critical.
4. Not affected by reactions to pretesting
An additional source of bias may be present when posttest results can be influenced by the results of a pretest. For instance:
- a physician might be influenced by a previous diagnosis or opinion made on a patient
- a participant taking the test the second time may be more prepared to the type of questions asked, etc.
So a study involving a pretest may be measuring the effect of the intervention along with this test-retest effect (a.k.a. sensitization to pretest). As this design does not have any pretest, it will not be subject to this bias.
5. Can be done when a pretest is not possible
This is especially useful when the measurement itself is very expensive either in terms of time or money, or when its results can be easily predicted or constant. For instance, a pretest is unnecessary if the outcome we want to measure is mortality and all participants are alive at the start of the study.
Limitations of the posttest-only control group design
1. High risk of attrition bias
This is due to participants quitting the study for different reasons between the study groups making the groups unequal anymore. Note that the risk of attrition bias is higher when the intervention takes a long period of time to be implemented (in the order of days or weeks).
The absence of a pretest makes it very hard to detect and control this bias.
2. The effect of the intervention on subgroups is not clear
In this posttest-only design we cannot compare the outcome with pretest measures, meaning that we cannot investigate which subgroup of participants responded more to the treatment or which subgroup did not respond well.
3. Requires a large sample size
The minimum sample size required for this study will be larger compared to studies without random assignment (i.e. quasi-experiments), this is because randomness can correct the differences between the treatment and control groups better as the sample gets larger.
4. Less generalizable than observational designs
Low external validity is a general characteristic of experiments for 2 reasons:
- participants who agreed to be part of the experiment may be different from those in the population on which we would like our results to generalize.
- experiments happen in a closed and controlled environment which is not always representative of a real world scenario.
Observational studies however, do not have such limitations as the investigator is just an observer of natural events, watching and recording them as they happen without controlling or influencing them.
In general, a highly controlled study will have a high internal validity (i.e. less bias) and a low external validity (i.e. low generalizability).
Example of a study that used a posttest-only control group design
In 1993, Topf and Davis used a posttest-only control group design to examine if CCU (Critical Care Unit) noise affects REM (Rapid Eye Movement) sleep.
So they randomly assigned 70 women with no hearing or sleeping problems to attempt to sleep in one of the following conditions:
- noisy environment (the subjects listened to an audiotape recording of CCU sounds): treatment group
- quiet environment: control group
Note that this experiment was done in a sleep laboratory.
Their results showed that CCU sounds can cause poorer REM sleep.
What can we learn from this example?
- The absence of a pretest was justified because participants had no sleeping problems before the experiment.
- This study may suffer from a generalizability issue because it used laboratory subjects that may not be representative of the entire population.
References
- Campbell DT, Stanley J. Experimental and Quasi-Experimental Designs for Research. 1st Edition. Cengage Learning; 1963.
- Shadish WR, Cook TD, Campbell DT. Experimental and Quasi-Experimental Designs for Generalized Causal Inference. 2nd Edition. Cengage Learning; 2001.