Dear Pratap Sing,

The importance of the two things "Explanation", and "Prediction" depends on the necessity of the user or the stakeholder associated. But, in most of the time, both of them are required in an analysis for better understanding the solution to the problem at hand.

Thanks,

Sobhan

In my opinion, explanation is the core. While you have a prediction without a proper explanation, your model would tends to fit a very narrow variety, and it would fail with a tiny change on the circumstances where the prediction is worked. Here, we shall mention something else. Statistics is a tool that to explain first, prediction MUST take place after exploring and understanding. But at the modern age, statistics almost always used for prediction without explanation.

I do not agree. Both are equally important but not at the same time. Attached is the classic work on the subject where the author discusses conditions where one is more important than the other. Notice that different methods of forming the model are used in each case.  For example variable selection is only used in prediction. Hope this helps, David

Explanation and prediction go hand in hand.  Explanation slightly precedes prediction because prediction is based on it. To that extent the former is more important than the latter.

I think both are equally important in the scentific aspect. Explanatation is a way to show that you have fully undertood the problem and especially you were not making any invalid assumptions when solving the problem. While on the other hand, you have to prove the effectiveness of your models via making predictions on unseen data (generlisation). 

I try to set out in this paper  under what conditions you get successful predictions by using critical realism and open and closed systems .- so it is not just about method but also the thing that you are studying and what you can do to it

https://www.researchgate.net/publication/256801634_The_practice_of_quantitative_methods

Chapter The practice of quantitative methods

The thought of model comes to mind first to explain the relationships that exist between different variables under consideration. If there is sufficient evidence of good relationship between dependent and independent variables then comes the question of predictability. For example, I want to see whether sales of cars in my area is any way related to Loan amount issued by the bank, Consumer price index and GDP. Apply the multiple regression model available on SPSS. Based on R^2 value, you can decide whether the model fitted is good or not OR the variables theoretically believed to be related, do really matter in the sales of the cars or not? A high R^2 (Significant also) will encourage me to use for the prediction of car sales for the year 2019 and 2020. So the explanation of relationship between different variables comes first and then if model is found to be very good, the question of prediction comes thereafter. If, before hand, you have sufficient evidence of relationship between dependent and independent variables then go directly for the development of a model which can be used only for prediction.

for me both are very important

Mavendra -

If your goal is to study the impact of various (independent) variables, that is one thing. But if you want the best prediction of y, that is another which could even employee principal components (which I don't think I can recommend, because even when your goal is prediction, not explanation, I'd like some idea as to what is involved).

In the example in the appendix of https://www.researchgate.net/publication/48178170_To_Explain_or_to_Predict, an underspecified model, though biased, may have better overall accuracy, but you would not know all the important variables. A larger sample size, I think, might change this, looking at what is said about that example.

Cheers - Jim

Dear Pratap Sing,

The importance of the two things "Explanation", and "Prediction" depends on the necessity of the user or the stakeholder associated. But, in most of the time, both of them are required in an analysis for better understanding the solution to the problem at hand.

Thanks,

Sobhan