The present paper seeks to cast scepticism on the validity and value of the results of all earlier studies in India on energy demand analysis and forecasting based on time series regression, on three grounds. (i) As these studies did not care for model adequacy diagnostic checking, indispensably required to verify the empirical validity of the residual whiteness assumptions underlying the very model, their results might be misleading. This criticism in fact applies to all regression analysis in general. (ii) As the time series regression approach of these studies did not account for possible non-stationarity (i.e., unit root integratedness) in the series, their significant results might be just the misleading result of spurious regression. They also failed to benefit from an analytical framework for a meaningful long-run equilibrium and short-run `causality' in a cointegrating space of error correction. (iii) These studies, by adopting a methodology suitable to a developed power system in advanced economies, sought to correlate the less correlatables in the context of an underdeveloped power system in a less developed economy. All explanations of association of electricity consumption in a hopeless situation of chronic shortage and unreliability with its generally accepted `causatives' (as in the developed systems) of population, per capita income, average revenue, etc., all in their aggregate time series, might not hold much water here. Our empirical results prove our secepticism at least in the context of Kerala power system. We find that the cost of dispensing with model adequacy diagnosis before accepting and interpreting the seemingly significant results is very high. We find that all the variables generally recognised for electricity demand analysis are non-stationary, I(1). We find that all the possible combinations of these I(1) variables fail to be explained in a cointegrating space and even their stationary growth rates remain unrelated in the Granger-`causality' sense