“In the left side table of contents, click on “Award Search and Proposal Deadline”.

w.

STEP 1 – Go to NSF Award Search Introduction webpage where 4 types of search are described. Two types of searches of immediate interest are: 1) the Program Information search (Figure 3) that includes Program Officer, and 2) the Awardee information search where you can query by Principal Investigator and limit search to Active Awards Only, Expired Awards Only, or Historical Awards.

https://www.fastlane.nsf.gov/NSFHelp/flashhelp/fastlane/FastLane_Help/fastlane_help.htm#introduction_to_fastlane.htm

STEP 2 – Click on “Search for Awards” near top of webpage, and go to NSF webpage that lists the steps to Search for Awards.

STEP 3 – Click on “URL” under “1.”, and go to Award Search webpage.

STEP 4 – To test the search function Award search webpage, under the default Awardee Information tab, under Principal Investigator I entered Michael Mann. I accepted the default “Active Awards Only” checked box at bottom of page. Then I clicked “Search”.

STEP 5 – The search results are then displayed at the bottom of the screen. For Michael Mann there are two awards listed, one for $100,000 and another for $299,761. The larger award is for “Analysis and Testing of Proxy-Based Climate Reconstructions” Award Number 0542356.

STEP 6 – In the left column, click on the Award Number 0542356. Viola! The Award Abstract page appears with the name of the Program Manager.

STEP 7 – Go to NSF Staff Directory search page. Type in name of Program Manager to get contact information.
http://www.nsf.gov/staff/

Here is a much better example of oscillation induced by seasonal displacement. The chart shows Sydney Australia daily TMAX for the 20 years to 2000, with a moving average of 365 days in black, which looks completely reasonable because all the seasons line up, and then at a moving average of 365+91=456, which oscillates like crazy. This is deliberately induced by using a roll which is one year plus one season. My point about diagnostics is that if such an oscillation is apparent when using a roll which is bounded by an annual unit (52 if by week, 365 if by day for annual, or 520 if by week, 3652 if by day for decades, etc) then that is a sure sign that the data is screwed up somewhere either by gaps, improper interpolation, or block displacement, or whatever.

This leads me to the opinion that working with daily data on annually bounded rolls is a lot safer than working with annual or monthly averages. It also has the side benefit of making it very easy to spot aberrant outliers by simply turning the roll off.

w.

Here’s one way to beat this NSF diversion and shirking of responsibility. Regardless of any hopes that NSF has for outside forces to self-police, the NSF has its own responsibilities to enforce compliance of NSF grant conditions for sharing data.

1. If someone wants data from a particular NSF-funded study, first find the name of the NSF project manager for the study.

2. Then, by certified registered mail/return receipt requested, send a request for data to the NSF-funded researcher with carbon copy to the NSF project manager.

3. If the NSF-funded researcher refuses to share the data, then, by certified registered mail/return receipt requested, send a letter to the NSF project manager providing documentation of the researcher’s refusal to share data and requesting that the project manager enforce compliance with NSF Grant General Condition on Sharing of Findings, Data, and Other Research Products.

4. If the NSF project manager does not enforce compliance, then ask the NSF project manager for the name of the NSF official that you can appeal this decision not to enforce compliance.

5. If notice that you intend to appeal does not cause the NSF project manager to reconsider the decision, then by certified registered mail/return receipt requested, send an letter to the NSF official appealing the decision of the project manager.

By the way, a similar case of shirking responsibility (passing the buck) is described by Willis at:

Climate Dynamics Passes the Buck

See especially Willis #4 response.

Apologies for being unduly terse.

The series under discussion is Indio Fire Station. Here is an annotated version of just that series. The large gap from 1982 to 1986 and the few smaller ones up to 1990 conspire to cause a seasonal displacement which takes 10 years to wash out.

The point of the original chart in #32 is to get a broad overview. In the case of California there are many regional climates, so the overall impression is a bit messy. Utah here is a bit more consistent.

Most of my work in time series is with longitudinal social studies such as voting intentions leading up to a major election. The data comes from surveys, and thus any one time series is connected to any other because a case is not just a single value, but a set of interrelated values which together describe a respondent. In this context the term ‘filter’ is never used as a synonym for smoothing. To filter a series means to remove cases on the basis of some Boolean or relational criteria, such as ‘Intend to vote Conservative filtered to Males with income over 100k’. The time series for this would comprise only those cases where Gender=Male is true, and, within that set, where Income greater than 100k is also true. The filter is in this context is some such expression as ‘Gender=Male AND Income.gt.100000′.

In my casual investigations of climate data I often use this sort of approach – for example, TMAX where TMAX is greater than 30C as a way of determining if heat waves are more or less prominent, or (TMAX+TMIN)/2 filtered to days where PRECIP is greater than 3mm, and so on. This is an excellent way to try to ferret out underlying patterns which would otherwise be totally obscured by a simple aggregate or annual average.

The standard smoothing algorithm for survey data is a moving average, and it seems to me to be a very useful way to get a good overview of underlying low frequency movements. I take UC’s point in #28 that a moving average on random data can be jiggered to produce just about anything you like (and I have witnessed this in practice many times) but if there is really a trend (and for climate that is nearly always true) then with clean and consistent data a long moving average should show it. Analytically, the usual practice is to use a linear regression on unsmoothed data to confirm the direction of the smoothed series, and then adjust the width of the moving average window according to the level of detail required.

However, temperature data sets are too often far from clean and consistent. One interesting application of the moving average algorithm is to diagnose chronologically displaced temperature data which has a seasonal signal. For example, if a block of winter temperatures are dated as being in a different season, then a moving average creates a visually obvious oscillation. Oscillation can also be caused by missing data, with the amplitude on rolled daily data being an indicator for how much an annual or monthly average will be distorted. See the top series in this chart between 1990 and 1996 for an example of oscillation induced by the patches of missing data from 1980 onwards.