Spatial Statistics
November 3, 2016
() Spatial Statistics November 3, 2016 1 / 31
Introduction
Spatial statistics are similar to traditional statistics, but they integrate spatial relationships into
the calculations.
() Spatial Statistics November 3, 2016 2 / 31
Spatial statistics will allow you to answer the following questions about your data:
How are the features distributed?
What is the pattern created by the features?
Where are the clusters?
How do patterns and clusters of different variables compare on one another?
What are the relationships between sets of features or values?
() Spatial Statistics November 3, 2016 3 / 31
Type of data analyzed
Location of crimes, animals, retail, industry, etc.
Land cover
Land use
Census/social science data
() Spatial Statistics November 3, 2016 4 / 31
Conceptual models
Inverse distance (spatial autocorrelation):all features influence all other features, but the closer
something is, the more influence it has.
Distance band:features outside a specified distance do not influence the features within the
area
Zone of indifference:combines inverse distance and distance band
() Spatial Statistics November 3, 2016 5 / 31
Conceptual models
K Nearest Neighbours:a specified number of neighbouring features are included in calculations
Polygon Contiguity : polygons that share an edge or node influence each other
Spatial weights : specified by user (ex. Travel times or distances)
() Spatial Statistics November 3, 2016 6 / 31
Identifying geographic distribution
Finding the centre of features
Mean center - Useful for comparing distributions of different features or over time
Central feature - Feature which is having the shortest total distance to all other features.
Useful for finding the most accessible feature
() Spatial Statistics November 3, 2016 7 / 31
Figure 1: Finding the center of the features
() Spatial Statistics November 3, 2016 8 / 31
Measuring compactness
Standard distance The extent to which the distance between the mean center and the features
vary from the average distance
The greater the standard distance, the more the distances vary from average - features are
more widely dispersed around the center
() Spatial Statistics November 3, 2016 9 / 31
In a spatial normal distribution,most features are concentrated in the center:
one standard deviation circle will cover about 68% of the features.
two standard deviation circle will cover about 95%
three standard deviation circle will cover about 99%
() Spatial Statistics November 3, 2016 10 / 31
Identifying patterns
Patterns are useful to:
Better understand geographic phenomena (ex. Habitats)
Monitor conditions (ex. Level of clustering)
Compare different sets of features (ex. Patterns of different types of crimes)
() Spatial Statistics November 3, 2016 11 / 31
Figure 2: Level of clustering of cut areas within a forest
() Spatial Statistics November 3, 2016 12 / 31
We can measure the pattern formed by the location of features or patterns of attribute values
associated with features (ex. median home value, percent female, etc.).
Figure 3: Patterns of different types of crimes
() Spatial Statistics November 3, 2016 13 / 31
Average nearest neighbour
Measures how similar the actual mean distance between locations is to the expected mean
distance for a random distribution
Measures clustering vs. dispersion of feature locations
Can be used to compare distributions to one another
() Spatial Statistics November 3, 2016 14 / 31
Identifying clusters
Global vs local Statistics
Global Statistics:Identify and measure the pattern of the entire study area. Do not indicate
where specific patterns occur
Local Statistics: Identify variation across the study area, focusing on individual features and
their relationships to nearby features (i.e. specific areas of clustering)
() Spatial Statistics November 3, 2016 15 / 31
Spatial autocorrelation
Spatial autocorrelation examines the spatial ordering of the geographic data. It measures whether
the pattern of feature values is clustered, dispersed, or [Link] addition, it explores the spatial
covariance structure in attribute data. Therefore spatial autocorrelation deals with both the
attributes and locations of spatial features. It indicates whether adjacent or neighbouring values in
the geographic data vary together and if so, how. It is affected by the scale of the spatial pattern.
In general, spatial autocorrelation is present when similar values cluster together on a map. There
are two commonly used measures of spatial autocorrelation: Geary’s index(c) and Moran’s index
(I).
() Spatial Statistics November 3, 2016 16 / 31
Geary’s index
This index was developed by Geary(1968)as a measure of spatial autocorrelation for area objects
with interval attributes. It is suitable measure for use in the analysis of data aggregated by
statistical reporting zones. This index measures the similarity of i’s and j’s attributes, cij which
can be calculated as
cij = (zi − zj )2
where zi and zj are the values of the attribute of interest for object i and object j. i,j are any two
objects measured on an interval scale.
A locational similarity measure wij was also used by Geary who defined it in a binary fashion with
wij =1 if i and j shared common boundary and wij = 0 is not.
() Spatial Statistics November 3, 2016 17 / 31
Geary’s index
Geary’s index is expressed as follows:
P P
i j wij cij
c= 2
P P
2σ i j wij
where σ 2 is the variance of the attribute z values.
If the value of c=1, the attributes are distributed independently of location. If the value of c < 1,
similar attributes coincide with similar locations. If the value of c > 1, attributes and locations
are dissimilar.
() Spatial Statistics November 3, 2016 18 / 31
Moran’s index
This measure was provided by Moran(1948), which has the advantage of giving a more logical
result with positive value implying that nearby areas trend to be similar in attributes, negative
value implying dissimilar. A zero value indicates uncorrelated, independent and random
arrangement of attribute [Link]’s index(I) also involves the use of cij and wij as in Geary’s
index and is defined as follows.
cij = (zi − z̄)(zj − z̄)
and
P P
i j wij cij
I =
s2
P P
i j wij
where s 2 denotes the sample variance.
The wij terms represent the spatial proximity of i and j and can be computed in any suitable way.
Both the indices as developed were for area objects, but they may be equally applied to points,
lines and raster objects provided an appropriate method can be developed to measure the spatial
proximity of a pair of objects. For point objects, one can compute the distances between pairs of
points and use inverse distance weighting to compute similarity. Another approach is to transform
the points into areas by partitioning the study area into Thiessen polygons. For line objects, if the
lines represent links between nodes with attributes, cij will measure the similarity of the attributes
of each pair of nodes while wij will measure the links between them.
() Spatial Statistics November 3, 2016 19 / 31
Figure 4: Computation of I values to test for statistically significant clustering
() Spatial Statistics November 3, 2016 20 / 31
Example of spatial autocorrelation
Spatial autocorrelation is typically applied to examine spatial distribution patterns of phenomena.
Can(1993) carried out a research on residential quality assessment in a city of New York. To
construct the residential quality score, he employed the following seven variables to capture the
demographic dimensions of socio-economic variations:
1 percentage of non-white persons
2 percentage of female headed single parent households with children under 18
3 percentage of occupied housing units that lack complete plumbing facilities
4 percentage of occupied housing units that have 1.01 or more persons in that room
5 percentage of vacant housing units
6 the median value of specified owner occupied housing units
7 the median contract rent of specified renter-occupied housing units.
These data were obtained at two spatial scales: census tracts and census block groups. The seven
variables at each spatial scale were subjected to factor analysis and standardized component
scores were obtained. The first component accounted for 75% for total variation at the census
tract level and 68% at the census block [Link] components were mapped which is shown in
the next figure.
() Spatial Statistics November 3, 2016 21 / 31
Figure 5: Spatial patterns of principal component scores of residential quality in the city of Syracuse, New York
at the census tract(left) and census block group(right) levels
() Spatial Statistics November 3, 2016 22 / 31
Visual inspection indicates a clustering of the densest shading(disadvantaged) from the city
centre to the south and east of the city. However an important question arises: Which spatial
level should be used for this type of analysis? To find the answer, Can calculated Moran’s I to
measure the extent of spatial clustering among census tracts and census block groups with
respect to residential quality scores determined by factor analysis. The results of Moran’s I for the
two spatial scales were 0.5101(with a sample variance of 0.0082205) for the census tract level and
0.7552(with a sample variance of 0.0023878) for the census block group level. The conclusion
was that the extent of clustering was much stronger at the census block group level than at the
census tract level. So the census block group should be used in the formation of neighbourhoods.
() Spatial Statistics November 3, 2016 23 / 31
Identifying relationships
Regression analysis
With other statistical tools we ask WHERE something is happening?
With Regression Analyses, we ask WHY something is happening.
Why are there places in the United States where people persistently die young? What might be
causing this?
Regression analysis allows us to model, examine and explore spatial [Link] also helps to
predict.
() Spatial Statistics November 3, 2016 24 / 31
Linear Regression
Used to analyze linear relationships among variables.
Linear relationships are positive or negative
Regression analyses attempt to demonstrate the degree to which one or more variables poten-
tially promote positive or negative change in another variable.
() Spatial Statistics November 3, 2016 25 / 31
Linear regression techniques
Ordinary Least Squares (OLS) is the best known technique and a good starting point for all
spatial regression analyses. This is Global model which provides one equation to represent the
entire dataset.
Geographically Weighted Regression (GWR): This is local model which fits a regression equa-
tion to every feature in the [Link] variation incorporated into the regression model
() Spatial Statistics November 3, 2016 26 / 31
Measurement scale in GIS
The attribute pertaining to spatial objects shown in a thematic map can be recorded by four
different levels of measurement. These four levels of measurement are:
Nominal Scale
Ordinal scale
Interval Scale
Ratio scale
() Spatial Statistics November 3, 2016 27 / 31
Nominal Scale:It is the simplest of all because it uses only names as labels. Thus a map of
agricultural regions will show such descriptive scales such as wheat region, corn region or cotton
region. Here the nominal scale is applied over polygons. It may also be applied in the lines such
as in the names and numbers of the highways and to points such as the names of the settlements
in a world map.
Ordinal Scale:This scale shows ordering or ranks. A map can show the rank of cities in a country
by size, or in a study of residential desirability, the rank of each city according to its quality in life.
Similarly a classification of roads into first, second and third class is very common on road maps.
Also countries in the world can be classified according to population size. Soils can be classified
according to drainage condition. Thus ordinal scale can be expressed in different ways and is
applicable to points, lines and polygons.
() Spatial Statistics November 3, 2016 28 / 31
Interval Scale:Interval scale does not have a natural zero and uses an arbitrary instead.
Temperatures in 0 F or 0 C are examples because temperatures in 0 F and 0 C use the freezing point
of water at 320 C and 00 C respectively as arbitrary origin. So a temperature of 400 F is 80 F above
320 F and a temperature of 200 F is 120 F below 320 F giving a difference in temperature of 200 F
becuase they are measured relative to 320 F. So a temperature of 400 F cannot be twice as hot as
200 F.
In other words, subtraction makes sense on interval scale but not multiplication because the value
is relative from an arbitrary origin. In topographic maps, we deal with terrain height which is
measured relative to MSL which is not an absolute origin. As a result, terrain elevations are
measured on an interval scale.
() Spatial Statistics November 3, 2016 29 / 31
Ratio Scale:Unlike the interval scale, a ratio scale makes use of an absolute zero. Weight,speed is
an example of ratio scale. If one person weighs 70 Kg and another weighs 35 kg, the first person
is two times heavier. Division makes sense in ratio scale.
() Spatial Statistics November 3, 2016 30 / 31
Measurement Scale
The impact of scales of measurement will be on cartographic representation and statistical data
analysis and spatial modelling in GIS. The general rule is to ensure that different scales are not
mixed up. However it is quiet common that after an overlay of two maps in ratio scale, the
resultant new map will be in ordinal scale. In cartographic represntation of attributes, the scale of
meaurement can be combined with the concept of continuous and discrete data.
() Spatial Statistics November 3, 2016 31 / 31