Scientific research is premised on the concept of validation and verifiability. Scientific conclusions must be able to be replicated in order to be considered reliable and effective observations of the world in which scientists and researchers operate. As a result, the objective validity of measurements, the ways in which research can be generalized, methods to reduce bias, the public nature of findings, and the ability to replicate research results are five critically important elements to the theory of research design.
First and foremost, all measurements taken during a study must be objective, valid, and reliable. Measurements are, by their very nature, inherently objective records of the data taken in a study. Without measurement objectivity, a given study is assured to be victim of a number of research design errors, such as statistical biases in the way the data is collected. Objective data, then, becomes a necessary requisite upon which all other portions of a study are based.
Specifically, objective data takes the form of pure measurements and direct observations of a given metric. This data, in other words, is not presented in a manner that runs any risk of interpretation. Moreover, there can be no implications mentioned with the objective data. Objective data can take the form of measurements such as a patient’s heart rate, the observable rate of decay of a deceased organism, or any number of other purely measurement-based data observations that exist in lieu of any sort of implied interpretation or understanding.
In scientific research, research must be able to be generalized in order to make the findings applicable to the scientific community as a whole. If the findings cannot be generalized to the broader topic and theme of the research, the study itself lacks utility for the scientific community overall. Indeed, the idea that research must be generalized in order to ensure that the sample group or population tested in the research design is representative and reflective of the broader population from which claims and findings will be applied. Thus, generalization is critical to scientific research in that without the ability to generalize one’s research, the research is applicable to only the immediate determined sample size of the population.
If research can be generalized to the population as a whole, the study then has applicability to target groups outside of the one immediately studied. This enables the findings of given research to apply to the broader population from which the studied sample was representing—in theory, a research design that has been designed, implemented, and executed properly will have findings from a sample that apply to the entire population. This, of course, raises the issue of the question of statistical significance and whether or not a sample population is, in fact, representative of the entire population. If, and only if, a studied sample population can be judged to be representative and reflective of the broader population will a research study have its findings generalized to the entire population. Thus, generalization offers a way for research findings to take on a very real sense of meaning and applicability to the wider population that the studied sample attempts to portray.
Bias is an unfortunate reality of scientific research. While it is effectively impossible to reduce bias in a study, it is possible to reduce instances of structural bias in a given study, as well as reduce the likelihood of bias in interpretation. Bias, moreover, can appear at any time and stage of research, all the way from the ways in which simple data is recorded, measured, and portrayed, to structural inequalities in the way in which a study is conducted, to biased sample populations that lack the ability to be generalized to the entire population as a whole. Bias can also occur in the interpretation and discussion of a given study, though this is less significant in terms of the negative impact that the bias can have on the overall study and its capacity to be replicated.
Specifically, techniques to reduce bias in scientific research center around attempting to distance the researcher from any all interaction with the actual objective data to the greatest extent possible. In other words, a research should be distant and objective as impossible when it comes to recording notable measurements and making other observations that should be otherwise completely objective and without interpretation or analysis (Jackson, 2013). Important techniques in reducing bias include ensuring proper selection of sample populations, consciously avoiding manipulation and selective recall of data or observable measurements, and avoiding confounding variables in a given study.
Selection bias is perhaps the most common issue that scientific research faces, and is the most easily treated. Selection bias occurs when a researcher attempts to design a study in which the studied sample is believed to be an accurate representation of the whole population, but in actuality is a biased sample that is heavily skewed away from the desired population. In other words, selection bias takes place when a studied population is not reflective of the wider population that the research designer is attempting to make conclusions for. Selection bias is best reduced through randomization of sample populations, as well as ensuring that the study’s participants are not limited to a particular demographic that may skew the sample. In essence, the principal method for avoiding selection bias is to randomize and gather as much information about the prospective study participants as possible, and guarantee that the population is as representative of the broader population as possible. Other techniques to reduce instances of information bias, which is bias that comes as a result of improper recording, include better planning and the implementation of redundancy systems in order to maintain research integrity. Interviewer bias can also be avoided if the researchers take care to use only standardized questions that are without any significant bias in their application.
It is critical that the findings of scientific research are made public and applicable to general population. Public research is able to be criticized by the general population at will and acts as a quality control element that private, internal research does not allow. Having research made public is an institutional guarantor of quality and offers other academics the opportunity to attempt to replicate results, discover and observe flaws in research skills and methods, and otherwise act in a fashion to assist in the critiques and analysis of a particular set of scientific research. In addition to acting as a sort of quality and peer-review, releasing research to the public has the arguably more important benefit of making sure that all research findings are able to be used by others in the scientific community. This means that, once results and methodology are published, others are able to use the new research as the basis for more advanced knowledge and complicated work. All studies are limited by inherent factors, whether lack of resources, limited sample size, or a host of other factors; public research enables individuals in other communities to embrace your research and act to improve on it.
Lastly, replication of results is of critical importance. Research is, at its core, an attempt to showcase an observable phenomena. Without the ability to replicate the results depicted in the study, the entire validity of the scientific finding is thrown into question. Replication not only guarantees validity of results, but is necessary in order for the scientific community to understand that the initial research has valid findings, that external variables and relationships can be addressed, and to form the basis of new research based on the ideas, methodology, and findings of the earlier study. Without the ability to replicate research results, scientific advancements would be extremely limited. As such, replication and reproducibility are critical elements to scientific research, and care must be taken to guarantee that these issues are addressed.
Non-experimental research design is a type of research design that avoids active manipulation of variables and populations. Non-experimental research designs usually takes the shape of correlation data studies, which are, at their core, observational studies made of existing datasets or groupings of data made without direct manipulation of study participants. In other words, non-experimental research design is a method of organizing and construction a study that does not have any change or manipulation of the independent variable by the person conducting the study and instead has the researcher merely observe the relationship between two existing variables. Thus, non-experimental research lacks the ability to determine causality, but nonetheless contributes greatly to understanding overarching relationships between variables.
Non-experimental research designs, moreover, have the added benefit of usually using far less resources than an experimental research design, while still achieving significant enough results. At the same time, however, they do lack the ability to show causality, which is an important aspect in understanding relationships between variables. Despite the somewhat more limited scope of their design, non-experimental research designs are highly valued scientific pursuits that are not inferior or more likely to result in incorrect findings than any other research design.
A quasi-experimental research design is a type of research design that functions in a similar way to pure experimental structured design, but lacks some important features. A quasi-experimental research design, unlike a non-experimental research design, does have an independent variable that is controlled, manipulated, or otherwise impacted by the actions of the researcher (Goba, 2011). However, a crucial difference exists between a quasi-experimental research design and that of a pure experiment—the quasi-experimental research design lacks control over some of the factors that play into the overall structure of the research design.
To illustrate these lack of controls, a quasi-experimental research design will feature a manipulated independent variable, but it will not have control over other factors. These other factors include, among others, selection bias and population sampling errors. In a quasi-experimental research design, the researcher does not or cannot randomize his studied population groups, meaning that the overall studied group has not been filtered for selection bias. As each participant does not have an equal, randomized chance of being chosen for a particular group of the study, the populations are incapable of being statistically random and thus safe for analysis.
However, these are not necessarily criticisms of the research design, as the unique nature of quasi-experimental research design offers researchers the ability to conduct research that other models would not permit. In many situations, particularly involving healthcare and the treatment of living humans, it may be impossible or impermissible to conduct a pure experimental design. As quasi-experimental research designs are aimed at understanding causal relationships, they are appropriate for situations where a pure experimental design cannot be implemented (Goba, 2011). While they lack the strong, effective, randomization as a control factor, quasi-experimental research designs are still effective in understanding and comprehending an overview of the causal relationship between different variables and population groups. Thus, despite their limitations in determining a pure and unbiased causal relationship, quasi-experimental research designs can be used to great effect, and are recommended for use in areas of scientific research where the pure experiment research design cannot be achieved for practical reasons.
Lastly, the true-experimental research design stands as the most scientifically rigorous of all of these methodologies. A true-experimental research design is one that maintains a control group and a manipulated experimental group, and where the researcher has the capacity to randomize the populations in each group in order to ensure that each population sample is sufficiently randomized in order to prevent selection bias and other complications that result from the problem of non-randomized sample groups. In terms of the benefits of using a true-experimental research design, it is widely accepted that only this particular model or method of research design has the procedural integrity to show a true causal relationship between a manipulated and controlled independent variable and the resulting dependent variable output.
The ability to showcase a true causal relationship is the primary strength of the true-experimental research design. The basic premise of the study is that each of the sample groups are randomized, meaning that each individual has an equal chance of being placed in either a testing or control group. Once this is determined and the sample groups and control groups have been constructed and assigned, an accurate true-experimental research design will then choose a single, active independent variable upon which the researchers will attempt to manipulate. The researchers, having manipulated that single variable, have then controlled for all other possible external and internal risk factors that would otherwise tamper with the results of the study (Luthans, 1982). Most importantly, true-experimental research designs are often used to attempt to disprove a hypothesis, as their ability to illustrate a real causal relationship is invaluable when it comes to disproving a stated relationship between two variables.
The benefits of a correlational research design are that the researchers can create a study that understands the relationship between two different variables, and how they interact. Importantly, these studies show only the relationship between the two variables and it must be noted that there is no proven causal link between two correlated variables; instead, these studies only show that a relationship exists. Other benefits include the fact that correlational studies are often far easier to conduct, and that real world limitations often prevent the use of a true-experimental research design. Thus, correlational studies allow for insight into the relationships between variables while at the same time making sure that practical constraints are mentioned (Kennedy-Clark, 2012). Disadvantages to this approach include the aforementioned fact that the study can only show there is a relationship; it cannot show the researcher conclusively that the actual nature of that relationship. Specifically, a correlational study does not show the researcher which of the variables is acting upon the other and causing the observed relationship. While correlational studies provide an interesting base upon which to conduct more research, they are in and of themselves not capable of proving a causal relationship between two variables.
The benefits and drawbacks of true-experimental research designs, however, are equally apparent. A true-experimental research design allows the researcher to search for and discover a true causal relationship between two variables in a controlled and manipulated fashion (Luthans, 1982). The fact that the research design allows the researcher to actively control all the constituent parts of his study allows him the opportunity to exclude all extraneous factors in order to enable him to show a true causal relationship. However, there are many drawbacks to the use of the true-experimental research design. Most notably, it can be said that true-experimental research designs do not reflect real world situations, and that the situations in which the researchers find their results can attempt to portray relationships in a vacuum. As a result, true-experimental research designs lack some of the real world applicability that other research designs may offer, and may suffer from less generalization capacity as other research designs. Lastly, true-experimental research designs can be expensive, difficult to construct, and may pose ethical and practical difficulties when it comes to the medical studies and other examples where human lives are at risk.
Goba, B., Balfour, R. J., & Nkambule, T. (2011). The nature of experimental and quasi-experimental research in postgraduate education research in South Africa: 1995-2004. South African Journal of Higher Education, 25(2), 269-286.
Jackson, M., & Cox, D. R. (2013). The principles of experimental design and their application in Sociology. Annual Review of Sociology, 39(1), 27-49. doi: 10.1146/annurev-soc-071811-145443
Kennedy-Clark, S., & Australian Association for Research in, E. (2012). Design research and the solo higher degree research student: strategies to embed trustworthiness and validity into the research design: Australian Association for Research in Education.
Luthans, F., & Davis, T. R. V. (1982). An idiographic approach to organizational behavior research: The use of single case experimental designs and direct measures. Academy of Management Review, 7(3), 380-391. doi: 10.5465/AMR.1982.4285328