Answer: The instrument is exhibiting a daily drift plus occasional noise spikes. First priority is to try to identify and remove the drift. Second priority is to minimize the disruptive effects of any residual drift. Possible sources of daily cycles are daily temperature fluctuations and voltage fluctuations; try changing these and alternative variables substantially while running the chart recorder.
If you can identify the variable, try to prevent it from affecting your measurements (e.g., voltage regulator), or quantify the relationship and monitor that variable during all measurements, so that you can apply a correction. If the cause of the daily variations is unknown or unmeasurable, choose an experimental design that minimizes its effect. The most obvious is to take either a zero reading or a calibration-standard measurement along with each widget measurement, depending on whether drift is in zeroing or in sensitivity, respectively.
The cause of the intermittent noise spikes is likely to be quite elusive. Because they are sudden and short-lived, they could make some measurements much less accurate than most, and they could affect only one of a paired measurement. One approach would be to measure the zero or standard both immediately before and immediately after the sample. If the two zero/standard measurements differ by more than a predetermined threshold, reject this measurement set and do another.
Computation and Information Handling
Computers are wonderful productivity enhancers. Whether for word processing, figure preparing, calculating, or extracting the most information from data, computers are essential to modern science. When I was a young scientist, I would give a draft manuscript to a secretary for typing, have one or at most two rounds of revisions, and submit it. I would give a roughed-out figure to a draftsperson, flag the most glaring drafting errors for revision, and submit it. Now I do my own typing and drafting, and I do dozens of revisions! The process as a whole may be slower, but the final product is certainly more polished.
Basic computer literacy for scientists includes proficiency in all of the following:
- an operating system (Windows®, Macintosh®, Unix®, or Linux®);
- word processing (e.g., Word® or Word Perfect®);
- spreadsheet analysis (e.g., Excel®); and
- a web browser (Netscape® or Internet Explorer®).
Most scientists also need one or more of the following:
- a graphics program (e.g., Kaleidagraph®);
- presentation software for slides and transparencies (e.g., PowerPoint®);
- image handling software (Photoshop® or Canvas®); and
- a statistical package (WinStat®, MINITAB®, SAS®, or SYSTAT®).
For some kinds of computation, speed is power. The current generation of computers is capable of solving more complex problems, involving more dimensions or variables, than were feasible even five years ago. The fastest vector machines, such as the Cray, are approaching their ultimate speed limits. Parallel processing, in contrast, is not bound by those limitations. Today’s largest computational tasks are massive because of the size of matrices or datasets, rather than because of the number of different kinds of computations. Such problems are well suited to parallel processing. The CM-2 Connection Machine, introduced in 1987, is an example of massive parallel processing: ef-
