08.02.2013 Views

New Statistical Algorithms for the Analysis of Mass - FU Berlin, FB MI ...

New Statistical Algorithms for the Analysis of Mass - FU Berlin, FB MI ...

New Statistical Algorithms for the Analysis of Mass - FU Berlin, FB MI ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

4.2. STATISTICAL REMARKS 69<br />

1. Validity: characterizes how well a measure reflects <strong>the</strong> system properties<br />

it was supposed to represent.<br />

2. Reliability (or precision): addresses <strong>the</strong> consistency or repeatability <strong>of</strong><br />

<strong>the</strong> measurement process.<br />

3. Amplitude: how well a measure represents abstract or higher order constructs<br />

and complex properties.<br />

4.2.2 Making Errors<br />

Measurements are generally made through <strong>the</strong> use <strong>of</strong> a measurement instrument,<br />

which is based on a scale that should have <strong>the</strong> same underlying relationships<br />

as <strong>the</strong> system property being measured. Formally put, a scale is a<br />

predefined mapping from one domain to ano<strong>the</strong>r (e.g. <strong>the</strong> volume expansion <strong>of</strong><br />

mercury to degrees <strong>of</strong> Celsius). The mappings can <strong>of</strong> course induce uncertainty<br />

e.g. through <strong>the</strong> use <strong>of</strong> fuzzy scales to represent <strong>the</strong> degree to which a property<br />

is considered present (Benoit and Foulloy, 2003). Obviously, <strong>the</strong> construction<br />

<strong>of</strong> a scale can be a source <strong>of</strong> error as well as each observation itself, which is<br />

a random variable with an underlying distribution (Potter, 2000). Basically,<br />

<strong>the</strong>re are four primary sources <strong>of</strong> measurement error:<br />

� Random error: non-deterministic variation from any source impacting<br />

<strong>the</strong> system including <strong>the</strong> system itself.<br />

� Systemic error: derives from construction <strong>of</strong> <strong>the</strong> measure or definition <strong>of</strong><br />

<strong>the</strong> measurement process and comes in <strong>for</strong>m <strong>of</strong> measurement bias.<br />

� Observational error: <strong>the</strong> oversight <strong>of</strong> key system properties requiring<br />

measurement or using <strong>the</strong> wrong measures <strong>for</strong> identified system properties.<br />

� Experimental error: <strong>the</strong> influence <strong>of</strong> environment conditions (such as<br />

temperature, air pressure, or response time <strong>of</strong> <strong>the</strong> operator) which affects<br />

<strong>the</strong> measurement process<br />

These errors create divergences between <strong>the</strong> perceived state <strong>of</strong> a system<br />

and <strong>the</strong> true state and can yield misleading insights and thus, must be addressed<br />

in any measurement framework. They will be part <strong>of</strong> <strong>the</strong> measurement<br />

process even when <strong>the</strong> system is welldefined (Krantz et al., 1971). Error is<br />

an inescapable feature <strong>of</strong> measurement (Mitchell, 2003) and can be partially<br />

addressed with statistical <strong>the</strong>ory as described in chapter 3.<br />

4.2.3 Experiment and Inference<br />

As mentioned in <strong>the</strong> previous sections when experiments are carried out, <strong>the</strong><br />

results <strong>of</strong> <strong>the</strong> measurements and <strong>the</strong> subsequent statistical analyses are <strong>of</strong>ten<br />

stated in <strong>the</strong> language <strong>of</strong> ma<strong>the</strong>matics or, more precisely, in that <strong>of</strong> <strong>the</strong> <strong>the</strong>ory<br />

<strong>of</strong> probability. These ma<strong>the</strong>matical statements have <strong>the</strong> beauty <strong>of</strong> being<br />

objective, precise and clear. On <strong>the</strong> o<strong>the</strong>r hand, this might induce people to<br />

hide inadequate experimentation behind a brilliant (ma<strong>the</strong>matical) facade. A<br />

fact, un<strong>for</strong>tunately <strong>of</strong>ten seen in scientific publications. Let discuss a simple<br />

example: data consisting <strong>of</strong> two columns <strong>of</strong> numbers, say x and y, can always

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!