Cluster Analysis

Hot and Cold Spot Clustering of Emergency calls in Battalion 2, and Household Income in Dallas County.

Problem Statement

After the Fort Worth Fire Department established that the emergency calls in Battalion 2 were clustering. It wanted to know where exactly this clustering was taking place. The chief of the department wanted to see the results displayed on the map as hot spots or cold spots. The department had no idea whether there is any relationship between the population census data and the incident calls. On the other hand, Dallas County Economic Development Office has Household Income data from the 2010 Census. But it has no idea where the high income and low income households are clustered, so that it can plan its charity collection efforts and job creation efforts, depending on the income brackets. The objective of this assignment is therefore to conduct a hotspot analysis in Battalion 2 and Dallas county, on incident calls and household incomes respectively.

Analysis procedure

To address the problem, I used ArcGIS Pro 3.0.2. I also used Cluster and Outlier Analysis (Anselin Local Moran's I) tool when doing a hot spot analysis for the emergency calls in Battalion 2. Data used were Calls for Service data that the fire department compiled for January 2015, census block-group-level data, and the street network data to give the map some reference. For the Dallas County income hotspot analysis, I used Hot Spot Analysis (Getis-Ord Gi*) tool. Data used were the census block-group-level data, and the street network data. All these datasets were provided by the course instructor.

In part one (a) of the assignment, I used Cluster and Outlier Analysis tool using inverse distance. I used Calls for Service layer which has data for January, 2015. The incident priority ranking (FEE) field was used as the input field, with zero permutations. This displayed results of Moran's I analysis which was symbolised on cluster and outlier type field, with areas hot spots, cold spots, and areas where outliers were clustering (mixture of high and low values). In part one (b), I rerun the same tool now using fixed distance band in order to control the search distances which the tool can use find a significant number of neighbours. The distance band was set to 900 based on the experience from previous assignment, where larger z-score was observed at 900ft. After creating a hot spot layer, I then symbolised the census block layer to graduating colors based on household income field. This was then used to compare with the hot spots layer in both parts to see if there is any observed relationships between the two layers.

In part two, I used the Hot Spot Analysis (Getis-Ord Gi*) tool to assess the Median Household Income local clustering patterns. The Input Feature Class was Census 2010 Income Dallas County layer, and this was set at fixed distant bands, which was estimated at 5280 feet. The output layer clear distinct cold spots (represented by the dark-blue) and the hot spots or high value areas which were represented by red color. The county could the concentrate its job creation efforts in the cold spots, and charity-collection efforts in the cold spots.

Process Diagram

Results

Application and Reflection

Problem Statement: With about over six million cases of malaria and 2,000 deaths reported each year, Malaria is one of the leading causes of morbidity and mortality in children under five. USAID/Malawi would like to help to reduce the prevalence and child mortality, but firstly has to establish whether the cases are just random of form a spatial pattern for targeted interventions. There has also been significant shortage of qualified health staff in Malawi to handle such disease burdens. Among the limited staff, there is an even distribution of staffing in most health facilities in Malawi. The shortage of well-trained, highly skilled, and equitably distributed health staff remains one of the most significant barriers to ensuring universal health coverage.

Data needed: Malaria incidence geocoded data. Health care staffing data.

Analysis procedure: I will use the Spatial Autocorrelation (Global Moran’s I) tool to assess whether the Malaria incidence cases in Malawi are random or form any clustering patterns. I will first assess the average square kilometers in Malawi villages (Administration boundary level 4) to determine the grid size to use. Using spatial join, I will join the grid data to malaria incidence health facility point dataset. Then I will run a definition query to make sure I rule out all the grids with no data. Finally I will run the Spatial Autocorrelation to determine the spatial pattern of the data. The data will be obtained from USAID/Malawi GIS database.

After determining that there is a clustered pattern, I will use the Cluster and Outlier Analysis (Anselin Local Moran's I) tool to conduct the hotspot analysis for the malaria incident data. I will use fixed distance band to control the distance at which the tool can search significant neighbours. This will help in informing the prevention efforts like spraying and distribution of mosquito nets in the hotspots. This will be coupled with pandemic control behaviour change messages in cold spots. I will also use Hot Spot Analysis (Getis-Ord Gi*) tool to conduct a hot spot analysis for number of qualified clinical staff in health facilities in Malawi. Malawi Health Facility Staffing layer will be used as input feature class layer, and the number of qualified clinical staff field will be used as target field. This will use fixed distance band as well. The results will be used for relocating staff from areas of high concentration to areas which have facilities with low number of qualified staff. This will also be weighted with catchment population before making any staffing recommendations.