Lecture 01.10.18
So… geography. Why is it so important?
Just like I mentioned in my last post, geography is a natural element of any and all analysis. Today’s lecture delved further into this idea to consider various issues arising from any study that uses geographic data, including:
- the modifiable areal unit problem (MAUP)
- the scale, grain and extent of a study
- the nature of the boundaries of a study area
- spatial dependence / heterogeneity.
While all of these issues are important for geospatial analysis, I will focus on what I found most relevant/interesting for the purposes of this blog post.
“There is no single natural scale at which ecological phenomena should be studied; systems generally show characteristic variability on a range of spatial, temporal, and organizational scales.”
(Leven, 1992)
When we are conducting a spatial analysis, how do we decide what scale to use? In landscape ecology, we might consider a particular animal’s typical habitat or range in order to identify an appropriate scale, called a characteristic scale. Sometimes, though, scale can be much more complicated to determine. Characteristic scales are not always knowable. If we truly want to develop a full understanding of a place in space (remember, context is key in any geospatial study), we should conduct a multi-scalar analysis.
We must also ask ourselves if the results of our analysis are dependent upon the spatial nature of the data. It may be that they are, instead, the results of a process – or some combination of both process and spatiality. For example, is the higher rate of heart disease deaths in southeastern US cities because people who live there are more likely to have unhealthy diets that contribute to heart disease? Or is it because are there are more people living in cities than rural areas? Or, is it because there are more hospitals and senior care facilities in urban areas, thus sick/old people are more likely to move there for care? We should always ask questions about the results of our analysis, and should never assume dependency on spatial analysis without looking at underlying processes (and vice versa).
The Modifiable Areal Unit Problem (MAUP)
The MAUP consists of two parts:
- What constitutes the object of spatial study?
Here again, there are issues of scale. The same set of data can give different results depending on the spatial resolution. This is called the scale effect. This can be seen in my lab assignment from GEOB 370, which compared the results of the same analysis conducted at two different spatial scales.
A larger scale (census tracts) generalizes the data over a wider spatial area. It loses more variability within in the patterns than when you use smaller dissemination areas. Same data, different results.
Moreover, there are various ways in which spatial units can be grouped at a given scale, which may lead to variability in statistical results (i.e. the aggregation or zoning effect). This is often seen in the United States gerrymandering, where political boundaries are redrawn in order to achieve desired election results.
2. Different areal arrangements of the same data produce different results, so we cannot say that the results are independent of the units being used. How can we make valid generalizations?
Issues arise also if we attempt to extend conclusions of one spatial resolution to another. For instance, because a spatial analysis found that a particular census tract earns a high average wage, it does not mean that every individual living in this census tract earns a high average wage. This assumption is called an ecological fallacy. This type of assumption also works in the reverse – the value for one individual cannot be applied to the entire coarser spatial region they occupy. This is called the individual fallacy.
These more general concepts overarch other important geospatial terms, including spatial autocorrelation, kriging, and Simpson’s paradox. Here, I’ll include some general definitions.
spatial autocorrelation: if a quantity in one spatial area makes its presence in neighbouring areas more or less likely, the phenomenon is spatially autocorrelated. This is often seen in biological phenomenon; as in, “the apple doesn’t fall far from the tree.” The dropping of apple seeds will result in a higher probability of apple trees growing nearby the existing tree.
kriging: a geostatistical method developed to model natural ‘random’ irregularities in sampled values. In GIS, kriging produces a surface that tells us how confident we can be with that surface. It is based on the concept of a regionalized variable that has three components: a structural (mean or constant trend), spatial correlation, and random noise.
Simpson’s Paradox: If the values of variables vary in correlation with each other, it might be impossible to obtain a reliable estimate of their ‘true’ correlation. For instance, areas with high unemployment rates are often associated with areas that also have high rates of other social-economic characteristics, such as a higher proportion of racial minorities, lower graduation rates, etc. Just how correlated are these variables? Often, the true answer is more complicated than just taking the result of our spatial analysis. We have to look deeper into each specific case/location/phenomena.