Statswork-Logo Statswork-Logo

Data Analysis, Results And Interpretation: Failure In Explaining The Causative Nature Between Variables

In practice, the data alone could not explain or infer something about the real problem; the common idea of the problem in mind is evaluated through the mode of data analysis. The though which we keep in mind is referred to as the causal inference about the problem. In this blog, we can know that what causality is it and where it results in failure in the statistical data analysis.

In every statistical analysis, the main interest is to frame a null hypothesis about the problem, and it always framed by the researcher and is tested whether it is true or not with the sample data. However, the causality is the concept where the problem can be handled with logical arguments without the data. The result cannot be compared with the statistical significance as we always do in any statistical practice. The concept is similar to the correlation technique, as this also identifies or make the researcher have an idea of the effect or cause of the relationship between the variables before the data collection (Dawid, 2004).

The common mistake in practice is that the researchers look for statistical information, understanding the correlation between the variables follows causational inference.  Many people say that correlation is the tool to identify the causal inference of the problems. However, in most cases, it not true. In addition, the correlation technique is used to study the linear association between the variables, and it involves the distribution of the observer variables, whereas, in the causal relationship, one cannot define the distribution. Bottom of Form

Causation is the relationship between the cause and effect, whereas correlation is a linear relationship between two variables. For example, suppose, if a doctor is interested in finding the effect of drug given to a patient, then he/she should understand the causality of the results. If the patient gets well, then the hypothesis could be whether this drug is the cause for the improvement in the patient or some other reason?. Here, the causal relationship cannot be measured, and it acts as a guide to frame the hypothesis regarding the problem. This can be tested by collecting the sample of data and tested with correlation analysis. Of course, the causal relationship can be tested not only through the correlation analysis alone, but it can also be tested using regression models (Johnson & Ahn, 2015).

Let us look at a few examples where the causational inference leads to invalid conclusions.

  1. Causal relationship in a reverse manner – Here, the correlation is obtained and concluded or interpreted in a reverse way. For instance, actually if A causes B, it may be concluded that B cause A in a reverse manner. A real-time example for this reverse inference is the working of a solar panel. We all know that solar panel generates power if the sun is visible for a longer time. But, it doesn’t mean that solar panel causes the longer time visibility of the sun.

  2. Hidden variable – Suppose we have hidden variables in the data, then the correlation could through an invalid result and contradicts the causal relationship. For example, consider the summer season, suppose an ice cream manufacturing company announced that there would be a 50% discount in the purchase of ice creams. It is obvious that people in the summer season will have ice creams to beat the sun, and it obviously makes the sales to go higher than usual. Suppose, if the summer season is ignored, then there is a causal relationship between the discount price and the ice cream sales. So, the summer season acts as a hidden variable here that causes an increase in sales (Pearl, 2009a).

  3. Coincidence – If there are two tasks that happened at the same time, then it doesn’t mean it is related. Suppose, assume that you have several computers connected to one server and suddenly you encounter a bug in your system, then it is not that the connected device causes it, and this can be from any other source.

  4. Bi-direction – When there are two things happening and have an effect on each other, then it is said to be bi-directional causation and lead to misinterpretation. For example, the number of thieves caught by the police is related to each other. At the same time, if the number of police increases, the number of thieves will decrease because they will get caught easily (Pearl, 2009b).

  5. Correlation is clear but not Causation – Consider the smoking and lung cancer study; it is easier to understand the correlation between smoking and lung cancer that smoking is the major cause of lung cancer. However, it doesn’t mean that smoking alone is a major part of causing lung cancer. Thus, in this case, the causal inference is not appropriate and differs from correlation.

Next, the common question is that does no correlation imply the absence of causation? The answer is no. It is not always true. Consider that you are driving a car, and if you press the accelerator, then the car goes fast, if you go upward in a mountain then you go slowly and maintain a constant speed. Suppose, if your friend drives the car and doesn’t know this information, then the mountain ride will be a disaster. Thus, if your friend knows the operation of the accelerator and how to ride in a mountain then it is easy to drive. Thus, in this case, the correlation between the people riding the car is different and results in no correlation. However, if both know how to drive in the mountain then there may be a causal relationship.

In conclusion, the statistical analysis is not enriched in identifying the causal relationship. In many practical situations, this often leads to misleading interpretations. Thus, care should be taken before conducting any statistical test and understand the cause of the problem and frame the suitable hypothesis (Pearl, 2009b).

Above blog Referred here

  1. Dawid, A.P. (2004). Probability, causality and the empirical world: a Bayes de Finetti Popper Borel synthesis. Statistical Science. [Online]. 19 (1). pp. 44–57. Available from: https://projecteuclid.org/euclid.ss/1089808272.

  2. Johnson, S.G.B. & Ahn, W. (2015). Causal Networks or Causal Islands? The Representation of Mechanisms and the Transitivity of Causal Judgment. Cognitive Science. [Online]. 39 (7). pp. 1468–1503. Available from: https://doi.wiley.com/10.1111/cogs.12213.

  3. Pearl, J. (2009a). Causal inference in statistics: An overview. Statistics surveys. [Online]. 3. pp. 96–146. Available from: https://projecteuclid.org/euclid.ssu/1255440554https://projecteuclid.org/euclid.ssu/1255440554.

  4. Pearl, J. (2009b). Causality. Causality: Models, Reasoning, and Inference. [Online]. Cambridge University Press. Available from: https://books.google.co.in/books?id=f4nuexsNVZIC.


jjgyou 1156131ghjh hkh21
jluj 484524

This will close in 0 seconds