Resa till USA: Dag 6

Day 6, 24th July  2012: To search and project

The ESRI Users Conference is so big that they have needed to create a lot of alternatives making it possible for oneself to focus on whatever special themes you are interested in. SInce the pedagogical aspects was (at least in part) covered på the preconference this weekend, I continued today to attend seminars about stuff I think I will have use for in my thesis, meaning modelling and analyzing spatial relationships, with special attention to finding patterns in data and determining how sure they are. This is done in part by classical means, applying statistical methods on the x- and y-variables and their values (called attributes), but also by using ways unique for 2D and 3D maps.

I have meantioned the methods, with names like Spatial Deviation Ellipse, a 2D way to show standard deviation, hot spot analysis, which finds which points of data is closer or further away then average, and cluster/outlier/grouping analysis. Suffice to say without being to technical, they are all very exiting (for a geonerd at least…).

They second part was leaving how and where behind and instead answering the question ”why?”. This is about understanding which variables are correlated, ie affects each other, about predicting future results, and to try understanding what the key factors are.

One thing which was mentioned several times was the potential for misinterpretations when just using your eye to understand map data, since symbols can be exaggregated or the number of colors affects to look of the map (where zones seems to exists).

I again got the six steps at the heart of this, called OLS regression, explained to me. Just to document this for my own sake in short form, they were:

1. check that variabel coefficients have the right sign

2. Check that the p-values of the coeffs are statistically significant (Koenker test => use robust)

3. check for variable redundancy (multilinearity), a VIF < 7.5 means your safe

4. Is there a model bias? The Jarqu-Bera test should be *not* stat significant

5. Model performance. Test different models and choose the one with the lowest AIC value (3 lower is stat sign.) and higher R2 value is best.

6. Spatial autocorrelation. The zones which are over/underpredicted should be random ie *not* stat.sign.

You see, that wasnt that hard! Just kidding, I’m berely hanging in there understanding the basics of those seminars, but damn was it interesting when coupled with whiteboards showing the results and the potential. ;-)

After OLS you can use Graphically Weighted Regression, GWR, which basically means that you study the maps visually and for example make it show which areas was most stat sign for a certain variable and act upon that. Say that in an area the number of car crashes was highly correlated to the number of road bumps. Then you examine how to change this.

I also went around in the Map gallery to explore all the fantastical (and some less so) maps. At the heart of geography the map still beats strongly.

En intressant sak i USA som ni antagligen känner till är att man här öppet dokumenterar folks ras, jag har för mig att man gör det på så sätt att man helt enkelt frågar folk om vilken de anser sig tillhöra. Detta i ett land som av frihetsskäl inte tex har ett nationellt leg eftersom det skulle kunna användas frihetsinskränkande.

Changes in etnic composition of the city areas

USA-map over the dominant races of each county (gray is caucasian, blua native american, green latino, red african, and so on)

Another differnce is the way sex offenders is openly shown and mapped, here in relation to schools

After the conference day was over I went to Mission Valley shopping center and bought some very wellpriced clothes and watched they Alien movie Prometheus. It was OK and not to scary. I would agree with the general critique that the presentation was very good but that the brain of the movie wasnt that high. Grade: Good, but barely so. SHould be seen at the cinema, though.

Map of the Jupiter moon of Io

GIS mapping is used to show that California can't bear inplanting a wolf population due to pop. density among other things. Something for Sala maybe?

Funny map done with serious mapping skills: what areas to avoid in case of a zombie plague

Mapping of burglare trends in a suburb

A 3D map of suicides in Grand Canyon. As usual I didnt see much of the third dimension with glasses on, due to my astigmatism I think.

Kitty duck tape. Way to go.

Is that...no it isn't...yep, it is "Hello Titty". Nothing for the younger ones, I guess...