My Brothers in Town

Baby Abigail at 5 months

My brother is in town today with the Swans  for the game against the Lions this weekend. We were able to get together at the Thai Rose restaurant at Woolloongabba. The Thai Rose was a good suggestion by Jason and Mel and John got to meet his grand-niece Abigail (now nearly 5 months) for the first time. We also enjoyed the food and the river-ferry ride back to John’s hotel afterwards.

SapidBlog has moved to WordPress.com

Owing to constant feature shinkage at Edublogs.org we have moved SapidBlog to a new home at WordPress.com. We were among the earliest supporters of Edublogs.org but of late they keep removing features from the free blog and putting them into the paid blog feature list. Eventually, we lost the capability to even save our blog to a backup  file without paying for it and decided that enough was enough.

 BBEQF8B3R6GF

Social Networking Safety

After my posting yesterday about the Facebook Troll that was charged, I thought that I should post more positive information today about social networking safety. There are some great rules for parents about socializing network usage by their teenage children at the US Federal Trade Commission (FTC) website: http://www.ftc.gov/bcp/edu/pubs/consumer/tech/tec14.shtm. The US FTC also run the Onguard Online safety site which features some interactive online games to help educate tweens and teens about their online safety.  These games seem both fun and challenging for 14-year olds and up. The game illustrated below is about  choosing your online friends safely. Tim (aged 16) thought it was “orright but pretty common knowledge.”  Give it a go and see what you think  (it might take a minute or so to load). There are numerous other interactive online quizzes availble at the  Onguard Online safety site for you to try as well: topics include phishing, identity theft, malware and spyware, laptop and wireless security and others (the “Friend Finder” interactive game will appear in the space below).

Facebook Troll Charged

Jessica Cook, 22, pleaded guilty earlier this week in Gympie Magistrates Court for helping to create a fake tribute page on Facebook for murdered Sunshine Coast woman Justine Jones and posting hate messages along with pornographic and offensive images (as reported in more detail in the Courier-Mail). The fake “tribute” page was created under the false name of Wendy Woods and she had apparently boasted about the media attention that she was receiving (as reported in The Gympie Times) .

The magistrate said that the case was “absolutely sickening” and warned that “heavy hand of the law will come down on you.” In this case the heavy hand of the law amounted to a  three-month suspended sentence,  a good behaviour bond and being banned from using social networking sites such as Facebook and Twitter. If she re-offends she’ll lose her $3000 bond.

Read more: http://www.news.com.au/technology/facebook-troll-jessica-cook-pleads-guilty-to-defacing-a-tribute-page-to-justine-jones/story-e6frfrnr-1225900174784#ixzz0yTxFsoVf

Importantly, this represents a warning to everyone using social media that they had better be aware of their personal safety while online because there are predators like this out there, all too ready to take advantage of you.

I think that this case also demonstrates that I wasn’t over-reacting when I deleted my Facebook site earlier this year or in making the numerous postings on Sapid blog concerning desecration of Facebook memorial pages:

http://sapidblog.wordpress.com/category/social-networking/facebook/

Facebook Defacer Faces Court

Disability pensioner, Bradley Paul Hampson, 29, of Tarragindi has faced the Brisbane Magistrates court over defaced Facebook tribute pages for deceased youngsters Elliot Fletcher (12) and Trinity Bates (8), that were defaced with pornographic images (more details in the Brisbane Times). He faces 5 charges over offensive use of the internet and possessing and distributing child exploitation material. He faces up to 10 years in prison if found guilty, according to a report in The Australian. He was denied bail.

This could be the first prosecution of its kind in Australia according to ABC Lateline – but about time in our opinion. At least now there is be some kind of legal deterrent for such offensive behaviour on social networking sites. It also establishes a will by the police to apply state and federal laws in virtual space as they would anywhere else. This we applaud.

Read-Write-Web have labelled the AFP as the "crocodile stasi" - click on the image above to read the original story.

However, the report in The Australian ["Facebook urged to ban teen tributes"] raises more worrying aspects to this case: apparently the Australian Federal Police (AFP) have urged Facebook to ban under-18s from posting memorial pages. This little attempted invasion of civil liberties can’t be applauded. Furthermore, it’s reported by the social media site Read-Write-Web ["Australian Police Deputize Facebook"] that the Australian Federal Police (AFP) have called for Facebook to hire a compliance officer and for Facebook to install a “rat button” on its pages in Australia that will allow reports to be sent the police with a click. This prompted Read-Write-Web to label the Australian police as the “crocodile stasi” and to suggest that:

Orwell may have been British but his nightmares must have wintered in Melbourne.

The Australian authorities have been slow to understand that heavy-handed attempts to control free access to the internet, be it by the Federal Police, or from Minister Stephen Conroy’s filter, presents a very negative image for Australia internationally.

You can follow all the articles on desecration of Facebook memorial sites on Sapidblog using this link:

http://sapidblog.wordpress.com/category/social-networking/facebook/

Doing Chemometrics (and Loving It?)

Well I’m not sure about the “loving it” part but this week the chemometrics sessions proceeded much more smoothly than the previous week. Setting up in class time was smoother and we didn’t create the expectation in students of being able to take the Matlab plug-ins home and expect them to work without trouble. This time we tried to create the expectation that the work was to be completed in the 3 hours of scheduled laboratory time. The use of the add-in outside of class is a bonus that shouldn’t be relied on.

This week we ran the new practical on determining octane numbers in fuels using NIRS (near-infrared reflectance spectrometry). This session is all computer-based because the data set used is from the paper: “Two data sets of near infrared spectra“ by John H. Kalivas in Chemometrics and Intelligent Laboratory Systems 37 (1997) 255-259. Nevertheless, it concerns a very practical question that everyone confronts whenever they refill their vehicle’s  fuel tank at a petrol bowser, namely, what are these octane numbers and why is one more expensive than another? 

In this practical the method of PLS (partial-least squares) was used to build a calibration model for a training set of fuel samples of octane number determined from the standard engine method.  The calibration model is then validated using samples of known octane number that are independent of the calibration model. PLS was carried out in an Excel® spreadsheet with a custom plug-in constructed using Matlab® EX builder for Microsoft Excel® as can be seen in the screen image below:

PLS plugin run from an Excel spreadsheet containing the fuel data for determining octane numbers

The screen above demonstrates the “variable selection” step where the blue diamonds show the root mean squared error for calibration samples (RMSEC) as a function of the number of latent variables in the model. In this case the RMSEC tends to underestimate the true error because of “overfitting.” On the other hand, the open-green diamonds are the RMSEP for validation samples which gives a better estimate of the true error and shows that the optimum number of latent variables to include in the PLS model is 3-5 (that is without excessively under- or over-fitting). The PLS plugin is also capable of performing cross-validation. Plotting the known versus estimated octane numbers for the validation samples is shown below:

Excel(R) plugin for PLS using for predicting octane numbers in fuels

The PLS model is able to successfully predict octane numbers of the published data set without evidence of significant bias and with a RSD of about 1%.

While there was a range of comments from the students it seemed like this approach of getting them immersed more in the experience of carrying-out chemometrics was really helpful. The capability of performing the PLS analysis in an Excel® spreadsheet is extremely important because it enables the students to start producing results almost immediately without having to learn the ins-and-outs of a more complex software application such as Matlab® or Unscrambler®.

If you would like to know more about the use of the PLS plugin described here then please leave a comment below or use the “Contact” page above.

Chemotherapy … NO! Chemometrics … YES!

Owing to retirements at the end of last year, this semester I’ve picked up lectures on the subject of “Chemometrics.” In previous years I’ve overheard students talking about going to do “chemo.” The joke being that chemometrics lectures were so difficult and painful that those students compared them to enduring “chemotherapy” treatments. Truth is that I enjoyed the joke too (too much) – but now the jokes on me – as I’m the one giving those same chemo lectures.

I decided that what had to be done was to make chemo more active and get students involved with learning by being immersed in doing practical activities. One set of activities started last week was a moderate success: the previous week students had brought in tap water samples from home and analysed them by our inductively-coupled plasma – mass spectrometer (ICP-MS). Last week students used those same data to perform chemometrics analysis of Brisbane waters using chemometrics techniques [specifically principal component analysis (PCA) and hierarchical cluster analysis (HCA)] using the Matlab® EX builder for Microsoft Excel® as shown in the diagram below:

Hierarchical cluster analysis (HCA) using Matlab EX builder add-in for Excel

In the above diagram the ICP-MS waters data is in the format of samples in rows and elemental results in columns. The HCA add-in consists of the above VBA dialog control which provides refedit control elements for data entry using intuitive point, select and click methods. Options for selection of distance metrics and linkage methods are provided using drop-down lists and numerical output is presented in the spreadsheet in the form of a distance matrix and a binary linkage tree. The resulting dendrogram is presented as a chart in Matlab format as shown in the diagram below:

Dendrogram for Brisbane water samples by ICP-MS

The Dendrogram above was obtained with standardised Euclidean distances and average linkages. It shows the tap water samples in red and the tank, reverse osmosis water, and bottled waters in black. The whole analysis is extremely rapid and simple to perform and only requires basic knowledge of Microsoft Excel®. ONCE IT IS SET UP!

The difficulty experienced was in setting up the add-ins an getting them to work reliably in a classroom situation. The add-in requires the Matlab compiler runtime (MCR) to be installed (this is essentially a stripped-down version of Matlab without any GUI, just the calculation engine). Then the add-in needs to be activated and running within Excel. The numerous steps involved all provide opportunities for things to go wrong and, unfortunately, they do go wrong quite often. So the students seem to appreciate the effort made in providing the software for them to use but have a fair bit of collective angst about getting it to work reliably. A lot of class time so far has been spent trouble shooting problems with the software rather than in actually doing chemometrics.

I’m embarassed to ask for help from the IT support people because I’m always working on such a last-minute basis. Also, I’m not sure that they’d like to see so much untested VBA code running around on their systems.

If you would like to know more about the use of the Matlab EX plugins described here then please leave a comment below or use the “Contact” page above.

Short-circuited Brains

In chemical analysis using instruments usually involves preparing a set of standard solutions, measuring the instrument response and preparing a calibration plot of concentration versus instrument reading. The concentrations for the samples (and their associated uncertainties) can then be calculated from the regression equation for the calibration. The required equations can be found in textbooks such as Daniel C. Harris, “Quantitative Chemical Analysis” and are relatively straight forward to calculate, though with a calculator, rather tedious and prone to error because of the numerous steps involved and the need to store intermediate values in calculator memory. Using a spreadsheet simplifies the procedure, but the built-in Excel function LINEST is difficult to use and regression function provided with the Data Analysis ToolPak in Excel lacks the ability to calculate a concentration x given a y from the instrument reading (or to estimate the uncertainty for the concentration x).

So for the past few years I’ve provided a spreadsheet template with macros, the latest version of which is shown below. Before the spreadsheet template (pre-2005) many students were submitting unsatisfactory reports that neither contained regression calculations nor provided uncertainties. Having spent much of the current weekend marking reports (yet again!) the students now find it easy to complete the regression calculations and uncertainties. But because they find it easy they also forget to engage the brain and think about what it is that they’re doing with this spreadsheet and reflect upon the significance of the results produced: I think that I’ve created a tool that makes it possible for some students to short-circuit their brains.

Calibration spreadsheet template

The problems that were evident included:

• thinking that it was satisfactory just to print the spreadsheet and attach it to their report without critical comment or analysis
• not discussing selection of fitting options such as “fit-through-the origin”
• not being aware of what the estimates of uncertainty were or how to use them in subsequent calculations

Note added in proof (22nd April): as I marked more reports from students, the situation wasn’t as bad as I first thought. Approximately 30% of students did make some brain-short-circuiting mistakes of the kind mentioned above. But another 30% did provide more in-depth treatment of their results with no indication of brain-short-circuiting and the remainder were somewhere in-between. The situation is clearly superior to that of pre-2005 because, I think, the spreadsheet template has assisted in making the requirements more explicit to students. Previously, the requirements were implicit, with statements of the kind: “use the regression equations given in the textbook” without stating how this was to be performed or what level of detailed analysis was expected.

One of the limitations of the current approach in providing the above regression calibration template to students is the issue of ownership: are students using it because they think they need to in order to gain a favorable grade? Do they use it as a useful labour and time saving tool that helps them get on with the important aspect of critical discussion of the results? In the past, a number of students have gone on to Honours, Masters or PhD work and have commented on how useful they found earlier versions of the template. Some previous students have used in in later year units. But usually students seem to forget about it at the end of the examinations (like so many things).

Next semester I intend to work further on the issue of ownership by offering students the choice of using a manual or scientific calculator, spreadsheet calculations using {=LINEST} (similarly to the textbook) and the calibration template method above. Students are then responsible for making their own choices. I’m confident that this will remove a further layer of brain-short-circuiting by students. Hope to make another blog entry about it then.

If you would like to know more about the use of the spreadsheet template described here then please leave a comment below or use the “Contact” page above.

Virtual versus Actual Desecration

This afternoon I heard Kelly Higgins-Devine talk on ABC radio Brisbane 612 about a case against 4 people accused of vandalising 82 gravestones in Toowong Cemetery being dismissed in a Magistrates Court because “it couldn’t be determined that the owners had not given permission for the headstones to be damaged.“

http://blogs.abc.net.au/queensland/2010/04/a-while-back-now-august-last-year-in-fact-you-might-remember-we-talked-about-dozens-and-dozens-of-headstones-being-damaged.html?site=brisbane&program=612_drive

Putting aside the obvious question about the impossibility of obtaining 82 denial-of-permission slips, presumably, from the surviving family members. The next question that occured to me: is there any connection between the desecration of physical graves and memorial Facebook pages to the recently deceased that have been discussed in Sapid Blog in recent posts?

Either in vandalising physical graves or in vandalising virtual memorial sites, its a small minority of delinquent buffoons that are acting in both cases: delinquent buffoons that in both cases need to be made aware of the consequences of their thoughtless and cruel actions by being brought to justice by the court system. But in both cases seem to be evading justice. One can only wonder what they’ll think they can get away with next?

I may be wrong, but its seems to me that our society has changed with regard to its permissiveness to such actions. Perhaps its the increasing materialism that we’re all exposed to that has made us think that spiritual world doesn’t matter so much anymore. Perhaps we are so busy keeping up with material possessions that that we are becoming oblivious to other more spiritual realities. Now that Facebook has brought a virtual dimension to millions of people’s lives: in the 3-realities of the work-a-day actual world; the virtual world of cyberspace; and the spiritual world; that the spiritual world is coming a poor last in many people’s lives.

So with our lives been spent in the enjoyment of the here and now it’s easily to forget about an afterlife and the need to respect the lives and memories of those deceased.

Microsoft Excel® ToolPak Confusion

Marking, marking, marking all last weekend. I was marking a practical reports where the students were required to estimate the concentration of an acidic product using the class data set from 19 sets of titrations with (i) phenolphthalein and (ii) methyl orange as indicators. They were then required to compare mean concentrations derived using a t-test and the precisions of the means using a F-test. Microsoft Excel® spreadsheets using the Data Analysis ToolPak® were used to perform the statistical tests.

F-test Results for Comparison of Acid Concentrations

Every student in the class (about 40 students) became confused in one way or another by the use of the F-test, Two-Sample for Variances – results are displayed above. It appears that this confusion is mainly generated by the ToolPak itself for some of the following reasons:

• only the one-tail probabilities are displayed (both one- and two-tailed probabilities are given with the corresponding t-test tool)
• the initially selected alpha value is not displayed in the output (and easily lost track of)
• it is not at all clear what affect setting an alpha value will have in the final analysis results

After much confusion myself, I can now clarify at least some of these issues. Firstly, the one-tailed probability can be converted into a two-tailed probability by multiplying by a factor of 2, i.e., the p-value for the above results is 0.046 x 2 = 0.092 (two-tailed). This was the main source of confusion for students since if p = 0.046 this is less than α = 0.05 and the null hypothesis is rejected (implying that the precisions of the class titrations by the two indicators are different at the 95% confidence level). However, in this case a two-tailed probability should be used and then we have p = 0.092 > α = 0.05 and the null hypothesis should be accepted (implying no difference between the precisions).

The other issue leading to student confusion was about what value to use in the dialog box under the Alpha: entry box (see above diagram). If it is left at the default entry of 0.05 then the F-value calculated is for a one-tailed distribution with α =0.05, F_crit = 2.22 which is less than F_test = 2.26 leading to the same result as above: rejecting the null hypothesis. However, to obtain the two-tailed value for F_crit we need to enter an Alpha: at half of 0.05 or 0.025. Then we obtain the two-tailed value for F_crit = 2.60 which is greater than F_test = 2.26 leading to the result that the null hypothesis should be accepted and that there is no difference in precision of the titrations for either phenolphthalein or methyl orange at the 95% confidence level.

Confused? Lost? Then you can understand the position that my students were in.

Given the stunning lack of success with my class I’m certainly tempted to not use the Excel® Data Analysis ToolPak ever again. But then again, since Excel® is a major tool used by industries that my students might find jobs with later on I’m pretty much stuck with it. At least for next semester I’ll have better instructions ready.

I’ve been using this practical excercise for 2 semesters a year since about 2004 and I’ve never encountered this situation before. It seems that this group is the most accomplished with performing titrations that I’ve ever encountered. No other class has ever come so close before to finding the indicator error from using these two indicators. In the past they found p-values that were always much greater than 0.05 and the issue of whether to perform a one-tailed or two-tailed test simply never arose.

I guess no good work ever goes unpunished.

---

The two-tailed test is most appropriate here since there is no a priori reason to confine the test to one side or the other of the distribution.