I am using Logistics performance index (LPI) for a panel analysis, however, index is issued ever second year. if i use the current value to fill the missing value of the previous year, could it be justified? Because LPI for 2018 is calculated based on information for 2917 & 2018, so does it make sense if I fill out the missing 2017 value with the available 2018 value? or do you recommend some other way of interpolation? Thank you!
If you have your predictive function f(x) = Y in whichever form with unknown variable X (independent variable). For each Xi, there is a corresponding outcome of Yi. For a missing Xk for year 2017, by substituting the value of X in 2016 and 2018, you can then use Y to obtain an estimate 2017. This value may be different than one obtained by substituting 2018 for 2017. For rationale and justification on how you treat missing data, at least you will have something to rely upon with methodological backing.
I would argue that what you suggest is not correct.
If for 2018 your equation looks like: LPI18 = f(X1,17, X1,18, X2,17, X2,18, ...) then you are including past values in the calculation. However, if for 2017 you also have LPI18 = f(X1,17, X1,18, X2,17, X2,18, ...), then you are including future values. In other words, you are using tomorrow, to predict today.
I would recommend that you simply run your model on data for every second year:
LPIt = f(X1,t-1, X1,t, X2,t-1, X2,t, ...).
Paul Louangrath yes, taking average of Xt-2 and Xt to obtain Xt-1, if I am not wrong, that's the most widely used way of interpolating. So, i think I will use it and see how does the results turn out.
Tomasz Marcin Napiórkowski what you said totally makes sense, I sawone paper that used KOF index of globalization. they used current value of index to fill for previous missing value to check the robustness of the model, but not for actual model.
I regret that I have to disagree with most of the answers, except the one from Tomasz Marcin Napiórkowski.
Interpolation is acceptable if the values used are known at the time of the interpolation.
Using the mean of X_t-2 and X_t to obtain X_t-1 cannot be correct as X_t was not in the information set at time t-1, that is, it was not known at time t-1.
It is preferable to use only previous values that were in the information set at the time of the interpolation, namely t-1, in order to replace the missing value X_t-1.
That means using X_t-2, X_t-3, ...and not X_t to interpolate X_t-1.
- Badges
- Science topic
- Similar topics
- Indexes