I am studying the article Barcelona baby boom: does sporting success affect birth rate?.

They are using month-data for their analysis of birth. They have normalised one month to be 30 days. So they have somehow cut values from one month to another.

There is a paragraph about removing abnormalities or probably better said gaps in the data

To evaluate a possible abnormality in a specific period, a single covariate (“intervention”) was included in the model under two different assumptions: (1) a one-time change modelled by using a covariate with zeros in all observations except a 1 in the pertinent month; and (2) a transitory change modelled by using a covariate with a value of 1 in the relevant month, which decreases exponentially over subsequent months by a factor of 0.7 (1, 0.71 , 0.72, and so on). The value of 0.7 was chosen by following standard recommendations.12 To estimate the percentage increase, we applied a logarithmic transformation to the series.

Assume that you have day data about the same event. Do you need to make such in manipulations to the data?

I may have misunderstood the purpose of the paragraph. If so, please, describe it to me.

  • $\begingroup$ (1) I do not think any data were removed. Although the text is vague, a natural interpretation is that birth counts were converted to rates per 30 days. (2) Although I cannot address the "need to" question, it is abundantly clear that such manipulations--in particular, basing the analysis on months--removes the statistical support from almost all the conclusions and claims the authors make. One could equally well conclude that births are caused by the mere anticipation of a sporting event or even--being more sceptical--that the analysis is worthless because it has no effective control. $\endgroup$
    – whuber
    Jan 6, 2014 at 15:41
  • $\begingroup$ @whuber Are these counts daily rates which they used to form monthly rates in the study? $\endgroup$ Jan 6, 2014 at 16:09
  • 1
    $\begingroup$ According to Figure 2, they are total births per month. Earlier the article states, "We looked at the number of births ... We analysed monthly birth data... ." But this is all beside the point: the article concerns a post hoc test of an anomaly that happened to be noticed. As with any such anomaly, such tests are meaningless. After all, I can look at any sufficiently long time series, model it, pick out the largest residual, and find any number of "explanations" in contemporaneous newpapers. By construction this residual will appear "significant." The whole exercise is pointless. $\endgroup$
    – whuber
    Jan 6, 2014 at 16:23


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.