Background
What should be done with missing data, or data gaps?  Researchers in climate change studies are constantly faced with this dilemma.  One common approach has been to produce a statistical estimate of the observed value.  Each estimate, however, has its advantages and disadvantages.  Another approach is to ignore the data gaps altogether, assuming the gaps are random events.  Such has been the policy of the USA's National Climatic Data Center, where monthly mean temperature values are calculated and printed in their official publication Climatological Data when data are missing for as many as 9 days in a given month.  What effect might this procedure have on the true monthly mean temperatures?

What was done
The authors investigated the effects of missing data on the calculated monthly mean maximum and minimum temperatures for 138 locations in the conterminous U.S.A. over the period 1951-1980.

What was learned
The effects of data gaps were shown to be most severe in winter, where one standard deviation from the true mean for monthly maximum and minimum temperatures ranged from less than "0.3°C for 2-day gaps to greater than 2.2°C for 10-day gaps."   The effects in the summer were less severe, ranging from "less than 0.1°C for 2-day gaps to greater than 0.7°C for 10-day gaps."  A spatial component was also noted, such that mean monthly maximum and minimum temperatures were affected more severely by missing data in interior continental locations than they were in coastal locations.

What it means
This study demonstrates the importance of using complete data sets in climate change studies.  The use of stations with as few as two missing days can introduce a departure from the true monthly maximum or minimum temperature on the order of 0.5°C.  The authors thus recommend "caution ? when monthly means are calculated with data gaps," and they note that such gaps in a data set "are especially a concern when they occur early or late in the record" in studies of climate trends.  We wonder what effect, if any, such data gaps have had on the reported 0.5°C rise in global temperature over the past century.