Plate 9.5: Gas Welders Working along Madzindadzi Road
3.9 Validity and Reliability
This section brings to the fore issues surrounding validity and reliability of data collected with the view to address the study objectives. The point of departure of spatial statistics is the presence of accurate location of units of analysis. Table 3.2 is a summary of descriptive data on level of precision in metres for all three sampling windows combined.
Table 3.2: Summary on GPS Precision Level (Field Experiences, 2017)
Descriptive Measure
Sample Size
Minimum Value
Maximum Value
Mean Lower Quartile
Median Upper Quartile
Standard Error
Statistic 642 6 98.4 13.5 9 11 16 0.25
It must be highlighted that the greater the number of metres, the lower the level of accuracy.
The results suggest that, on average, the location of all units of analysis was captured at a range of 13.5 metres and 75% of spatial points fell in a range of 16 metres. This is a true reflection that the accuracy level was very high. Taking the sample size in account, a standard error of only 0.25 metres was experienced implying that the locations of manufacturers deviated from the mean by 0.25 metres. Presence of outliers slightly stretched the mean upwards. Had it not been that, the average GPS precision level could have been lower than 13.5 metres.
However, the precision level was generally high regardless of the present outliers. The normalised collective precision level for the three study sites had a maximum value less than 30 metres. In this case, outliers describe very low precision level not compatible with other levels of precision in the data set. Perhaps, the following questions regarding precision level are worth to answer: are these outliers spatially distributed? Were the outliers experienced on selected days of the survey? These two questions are best answered by a detailed analysis of
79
precision level per site per day. As explained earlier in this chapter, Gazaland, Siyaso and the Complex were surveyed on days one, two and three respectively. The Mosaic Plot in Figure 3.6 provides the per-site GPS precision level and the percentage share of individual sites at different levels of accuracy. In the diagram, the three sampling windows, the Complex, Gazaland and Siyaso are represented by blue, brown and orange respectively. The horizontal axis gives different levels of accuracy ranging from 7 to 25 metres, and the vertical axis shows the percentage contribution by sampling windows at different levels of accuracy.
Figure 3.6: Home Industry by GPS level of precision (Field Experiences, 2017)
The thickness of bars in the Mosaic Plot describes the dominance in precision level. Figure 3.6 suggests that 7 to 13 metres ranges are dominant. Very few units of analysis were captured in the range above 13 metres as signified by the very thin bars. The column titled missing describes some results without geographical location. In spatial point analysis, such data is regarded as redundant. To enable further analysis, all manufacturers’ results without geographical location were deliberately omitted from the database at the data reduction stage.
Only usable data (that is, data with corresponding spatial location) was considered for further analysis. Though the percentage of such cases were insignificant (thin bar), it partly
80
compromised the results of the study because the concept of census blocks (Baddeley, 2010) could not be realised, that is mapping manufacturers in three sampling windows in their entirety. To complement this explanation and further clarify per-site GPS level of accuracy is the per-site, per-day GPS precision level indicated in Figure 3.7.
Figure 3.7: Per-Site, Per-Day GPS Precision Level (Field Experiences, 2017)
The set of Box and Whisker Plots in Figure 3.7 suggest that for all the three sites, 75% of study sites’ geographical location were very accurately captured with a series of less than 20 metres.
Extreme outliers were experienced on day one of the survey (in Gazaland) with a maximum range being 98.4 metres. After the technical facilitator attended the problem experienced on the first day of the survey, reasonable outliers of below 40 metres were documented on the second day of the survey (in Siyaso).
After critically studying the causes of outliers and taking corrective measures, all spatial locations captured on day three were devoid of outliers. Outliers were caused by capturing geographical coordinates under roof for manufacturers who work in closed spaces and some concerns in the underlying infrastructure of the mobile geo-application. This challenge was rectified by capturing the spatial coordinates outside buildings. It is critical to highlight that
81
only two enumerators had serious challenges in recording the spatial location of respondents in the first day of the survey. Figure 3.8 is a rough visual impression of per-site, per enumerator GPS precision level.
Figure 3.8: Per-Site, Per-Enumerator GPS Precision Level (Field Experiences, 2017)
The heat map in Figure 3.8 shows that in the first day of the survey two enumerators’ data collection devices reported geographic coordinates which had a lower accuracy degree. Level of accuracy was largely compromised by a high range of 98.4 metres which subsequently increased the total variance and standard deviation. The technical assistant fixed this challenge during the lunch break of day one. The level of accuracy then improved to at most 40 metres whilst using the mobile devices offline. Further clarifying issues surrounding reliability and validity of the findings of the study is the section on ethical considerations.