Some frequent ways of unwillingly misrepresenting experimental results

Many students and some researchers are ignorant of the fact that any of the following practices are statistically invalid and could be considered to be ‘research-results manipulation’ (=cheating):

  1. Repeating an experiment until the p-value becomes significant.
  2. Reporting only a ‘typical’ (=nice-looking) replication of the experiment, and presenting statistics (tests of significance and/or parameter estimates such as means and standard errors) based only on this subset of the data.
  3. Presenting a subset of the data chosen using a subjective criterion.
  4. Not reporting that outliers have been removed from the data presented or used in analyses.

A limitation in the current implementation of spectral objects

I noticed this morning an unexpected behaviour of .spct objects when manipulated using data.table syntax… they become data.table objects! Until I solve this problem please do not use data.table syntax for manipulating .spct objects. I will upload a bugfix release of package photobiology, after I make sure that all functions defined in the package behave as expected. This will happen later today or tomorrow.

[edited @ 14:14, 22 July 2014] This time the bug was not in the code I had written, but rather in the data.table package. The code has recently been corrected in the development version of data.table (1.9.3), but this version is not yet available through CRAN… I will most likely upload a binary archive of data.table 1.9.3 to the r4photo repository later today, and a new version of package photobiology marked as requiring this or later versions of data.table.