SHARE THE ARTICLE ON
As you strive to uncover causal (cause-and-effect) relationships between variables, you may often encounter ethical or practical constraints while conducting controlled experiments. Quasi-experimental design steps in as a powerful alternative that helps you overcome these challenges and offer valuable insights.
In this blog, we’ll look into its characteristics, examples, types, and how it differs from true-experimental research design. The purpose of this blog is to understand how this research methodology bridges the gap between a fully controlled experiment and a purely observational study.
A quasi-experimental design is pretty much different from an experimental design, except for the fact that they both manifest the cause-effect relationship between the independent and dependent variables. So how is quasi-experimental design different?
Well, unlike experimental design, quasi-experiments do not include random assignments of participants meaning, the participants are placed in the experimental groups based on some of the other criteria. Let us take a deeper look into how quasi-experimental design works.
Experimental design has three characteristics:
Manipulation simply means, evaluating the effect of the independent variable on the dependent variable.
Example: A chocolate and a crying child.
Here, the Independent variable is: the type of chocolate.
And dependent variable is: the child is crying for a chocolate.
So manipulation means the effect of an independent variable that is chocolate, on the dependent variable, that is the crying child. In short, you are using an outside source on the dependent variable. This proves that after getting the chocolate (independent variable), the child stops crying (dependent variable).
Randomization means sudden selection without any plan. Example: A lottery system. The lottery numbers are announced at random so everyone who buys a lottery has an equal chance. Hence, it means you select a sample without any plan and everyone has an equal chance of getting into any one of the experimental groups.
It means using a control group in the experiment. In this group, researchers keep the independent variable constant. This control group is then compared to a treatment group, where the researchers have changed the independent variable. Well, for obvious reasons, researchers are interested in the treatment group more as it has a scope of change of the dependent variable.
Example: You want to find out if the workers work more efficiently if there is a pay rise.
Here, you will put certain workers in the treatment group and certain in the control group.
Treatment group: You pay more to the workers
Control group: You don’t pay any extra to the workers and things remain the same.
By comparing these two groups, you understand that the workers who got paid more worked more efficiently than the workers who didn’t.
As for the quasi-experimental design, the manipulation characteristic of the true experiment remains the same. But randomization or control characteristics are present in contrast to each other or none at all.
Hence, these experiments are conducted where random selection is difficult or even impossible. The quasi-experiment does not include random assignment as the independent variable is manipulated before the measurement of the dependent variable.
See Voxco survey software in action with a Free demo.
The above description is overwhelming? Don’t worry. Here is the straight difference between the quasi-experiments and true experiments just so you can understand how both vary from each other.
Participants are assigned randomly to the experimental groups.
Participants and not randomly assigned to the experimental groups.
Participants have an equal chance of getting into any of the experimental groups.
Participants are categorized and then put into a respective experimental group.
Researchers design the treatment participants will go through.
Researchers do not design a treatment.
There are no various groups of treatments.
Researchers study the existing groups of treatments received.
Includes control groups and treatment groups.
Does not necessarily require control groups, apart from the fact they are generally used.
It does not include a pre-test.
It includes a pre-test.
All the explaining isn’t always enough, right? So we got it covered with clear examples that will help you set quasi-experiments and true experiments apart from each other;
Let us say you want to study the effect of junk food on obese people.
While starting the true experiment, you assign some participants in the treatment group where they are fed only junk food. While the other half of the participants go to the control group, where they have their regular ongoing diet (standard course).
You decide to take obese people’s reports every day after their meals to note down their health and discomfort if any.
Although, participants who are assigned to the treatment group would not like to change their diet to complete junk food for personal reasons. In this case, you cannot conduct a true experiment against their will. This is when quasi-experiment comes in.
While talking to the participants you find out that some of the participants want to try the junk food effect while the others don’t want to experiment with their diet and choose to stick with a regular diet.
You can now assign already existing groups to the participants according to their choices. Study how the regular consumption of junk food affects the obese from that group.
Here, you did not assign groups to the random participants and can be confident about the difference occurring due to the conducted experiment.
The advantages of a quasi-experimental design include
The disadvantages of a quasi-experimental design include:
Amongst all the various types of quasi-experimental design, let us first get to know two main types of quasi-experimental design:
You can picture non-equivalent group designs as a mixture of both true experimental design as well as quasi-experimental design. The reason is, that it uses both their qualities. Like a true experiment, NEGD uses the pre-existing groups that we feel are similar, namely treatment and control groups. But it lacks the randomization characteristic of a quasi-experiment.
While grouping, researchers see to it that they are not influenced by any third variables or confounding variables. Hence, the groups are as similar as possible. Like talking about the political study, we might select groups that are more similar to each other.
Let us understand it with an example:
Take the previous example where you studied whether the workers work more efficiently if there is a pay rise.
You give a pre-test to the workers in one company while their pay is normal. Then you put them under the treatment group where they work and their pay is being increased. After the experiment, you take their post-test about their experience and attitude towards their work.
Later, you give the same pre-test to the workers from a similar company and put them in a control group where their pay is not raised, and then conduct a post-test.
Hence, the Non-equivalent design has the name to remind us that the groups are not equivalent and are not assigned on a random practice.
Regression discontinuity design or RDD is a quasi-experimental design technique that computes the influence of a treatment or intervention. It does so by using a mechanism that assigns the treatment based on eligibility known as a “cut-off”.
So the participants above the cut-off get to be in a treatment group and those below the cut-off doesn’t. Although the difference between these two groups is negligible.
Let’s take a look at an example:
A school wants to grant a $50 scholarship to students depending on an independent test taken measuring their intellect and household.
Those who pass the test will get a scholarship. However, the students who are just below the cut-off and those just above it can be considered similar. We can say the differences in their scores occurred randomly. Hence you can keep on studying both groups to get a long-term outcome.
Apart from the above-mentioned types, there are other equally important quasi-experimental designs that have different applications depending on their characteristics and their respective design notations.
Let’s take a look at all of them in detail:
The proxy pre-test design works the same as a typical pre-test and post-test design. Except, the pre-test here is conducted AFTER the treatment is given. Got confused? How is it pre-test if it is conducted after? Well, the keyword here is “proxy”. These proxy variables tell where the groups would have been in the pre-test.
You ask the group after their program about how they’d have answered the same questions before their treatment. Although, this technique is not very reliable as we cannot expect the participants to remember how they felt a long time ago, and we surely cannot tell if they are faking their answers.
As this design is highly not recommended, you can use this under some unavoidable circumstances like the treatment has already begun and you couldn’t take the pre-test.
In such cases, this approach will help rather than depending totally on the post-test.
Example: You want to study the workers’ performance after the pay rise. But you were called to do the pre-test after the program had started. In that case, you will have to take the post-test and study a proxy variable such as productivity from the time before the program and after the program
This technique also works on the pre-test and post-test designs. The difference is, that the participants you used for the pre-test won’t be the same for the post-test.
Example: You want to study the client satisfaction of two similar companies. You take one for the treatment and the other for the control. Let’s say you conducted a pre-test in both the companies at the same time and then begin your experiment.
After a while, when the program is complete, you go to take a post-test. Now, the set of clients you take in for the test is going to be different than the pre-test ones, the reason being clients change after the course of the period.
In this case, you cannot derive one-to-one results, but you can tell the average client satisfaction in both companies.
The double pre-test design is a very robust quasi-experimental design designed to rule out the internal validity problem we had with the non-equivalent design. It has two pre-tests before the program. It is when the two groups are progressing at a different pace, that you should change from pre-test 1 to pre-test 2.
Due to the benefit of two pre-tests, you can determine the null case scenario. It assumes the difference between the scores in pre-test and post-test is due to random chance as it doesn’t allow one person to take the pre-test twice.
In the switching replications design, as the name suggests, the role of the group is switched. It follows the same treatment-control group pattern, except it has two phases.
Phase 1: Both the groups are pre-tested, then they undergo their respective program. Later they are post-tested.
Phase 2: In this phase, an original treatment group is now a control group, and an original control group is now a treatment group.
The main benefit it provides by inculcating this design is that it not only proves strong against internal validation but as well as external validation as well. The reason being two parallel conducted implementations of the program allows all the participants to experience the program, making it ethically strong as well.
NEDV design, in its simplest form, is not the most reliable one and does not work wonders against internal validity either. But then what is the use of NEDV?
Well, sometimes the treatment group may be affected by some external factors. Hence, there are two pre and post-tests applied to the participants, one regarding the treatment itself and the other regarding that external variable.
Wait, how about we take an example to understand this?
Let us say you started a program for testing history teaching techniques. You design standards test for history (treatment group) as well as showing historical movies (external variable). Later in the post-tests, you find out that along with the history scores, students’ interest in historical movies has also increased, suggesting that showing historical movies have influenced students to study the subject.
RPD design is used when the measures for the already existing groups are available and can be compared with the treatment group. The treatment group is the only group present and both pre-test and post-tests are conducted.
This method is widely beneficial for the larger groups per se; communities or companies. RPD works by comparing a single program unit with a larger comparison unit.
Let us make it clear with an example:
Consider a community-based COVID awareness program. It is decided to start the initiative in a particular town of a vast district. The representatives forecast the active cases in that town and use the remaining towns as a comparison. Now rather than giving the average for the rest of the towns’ COVID cases, they show their count.
All that studying but shouldn’t you know when to perfectly use quasi-experiments? Well, now as we are to the end of the matter, let us discuss when to use quasi-experiments and for what reasons.
Remember when we discussed the “willingness” of obese people to participate in the experiment? That is when ethics start to matter. You cannot go on putting random participants under treatments as you do with true experiments.
Especially when it directly affects the participants’ lives. One of the best examples is Oregon Health Study where health insurance is given to certain people while others were restricted from it.
True experiments, despite having higher internal validity, can be expensive. Also, it requires enough participants so that the true experiment can be justified. Unlike that, in quasi-experiment, you can use the already gathered data.
The data is collected and paid by some strong entity, say the government, and you use that to study your questions.
Well, that concludes our guide. If you’re looking for extensive research tools, Voxco offers a complete market research tool kit that includes market research trends, a guide to online surveys, an agile market research guide, and 5 market research templates.
Quasi-experimental design has a unique approach that allows you to uncover causal relationship between variables when controlled experiments are not feasible or ethical. While it may not posses the level of control and randomization that you have when performing true-experiment; quasi-experimental research design enables you to make meaningful contribution by providing valuable insights to various fields.
Let us say you want to study the effect of eating cheese on bad breath. So you make the people with not so bad breath take the treatment and the other half with bad breath to be in the control group. After taking the post-test you discover that the participants in the treatment group start to have bad breath.
The quasi-experimental are used to evaluate interventions without using randomization. It also interprets the problems using pre-intervention and post-intervention measurements along with non-random assignments.
A true experiment uses random assignment of the participants while quasi-experiments does not. This allows its wide use in ethical problems.
Quasi-experiments allots the participants based on a study, unlike true experiments where they have an equal chance of getting into any of the groups.
Quasi-experiment also makes use of the pre-test as well as post-test measurements which opens a door to before-after comparisons.
The quasi-experimental design does not randomly assign groups to the participants, rather it studies their nature and then treats them accordingly.
It studies the participants before and after the program known as pre-test and post-test which helps get an idea about the progress of the groups.
Quasi-experiments also are ethical, due to their non-randomization characteristic.
Quasi design or quasi-experimental design mostly resembles the true experimental design, just minus the key component. That is a random assignment.
Two prime quasi-experimental methods include:
Some other, rather equally important Quasi Designs are:
Customer touchpoints : What is it , how to identify them , it’s phases and examples SHARE THE ARTICLE ON Share on facebook Share on
Ethical Considerations in Predictive Analytics: Privacy and Bias SHARE THE ARTICLE ON Table of Contents Introduction Predictive analytics, a powerful data-driven tool, has altered decision-making