*4.1. Descriptive Statistics*

Table 2 shows that only 25% and 20% of respondents were afraid to cross the intersection and felt like making a detour to avoid crossing intersection. Based on this, only 52 respondents (approximately 28%) were defined as people who felt FoC at the intersections. Considering the increasingly degraded security situation throughout the country [45] and UN-Habitat's report that found that 89% of people in Yaoundé considered their neighbourhood to be unsafe [1] (p. 91), the low percentage of people reporting FoC was well below our expectations. However, the relatively small percentage obtained could be explained by the fact that our survey was conducted in a public place. Most of the persons surveyed were at the intersections in question, and felt more or less comfortable using these spaces.

Table 3 describes the frequencies of reporting FoC and the pedestrian density by intersection. The result of Fisher's exact test showed that the proportion of people who felt FoC significantly varied across intersections at 0.01 probability level. However, the relationship between the proportion of people who felt FoC and the pedestrian density of intersection was not significantly associated (*p* = 0.104) although they showed quite high positive correlation (0.800). Thus, we need to test this hypothetical relationship that 'high pedestrian density increases FoC at a congested intersection with higher density than a certain threshold level' after controlling for other factors.


**Table 3.** Fear of crime and pedestrian density by intersection.

Note: The proportion of people who felt FoC significantly varied across intersections at 0.01 probability level (results of Fisher's exact test). The proportion of people who felt FoC and the pedestrian density of intersection was positively associated (Pearson correlation = 0.800), but the significance level was marginal (*p* = 0.104).

### *4.2. Results of Multi-Level Binary Logistic Regression Analysis*

Table 4 presents the results of multi-level binary logistic regression models of FoC. Model 1 is an intercept-only (unconditional) model and Models 2 and 3 are random intercept models with di fferent set of independent variables.


**Table 4.** Multi-level binary logistic regression models of 'fear of crime'.

Note: Number of observations = 185; number of clusters = 5; \*\*\* *p* < 0.01; \*\* *p* < 0.05; \* *p* < 0.10; ICC: intraclass correlation; In multi-level logit models, level-1 variance is assumed to be π2/3 [57].

In the intercept-only model result, intraclass correlation (ICC) was 0.113. This means that 11.3 percent of the probability of feeling FoC is attributable to di fferences between intersections. However, the insignificant level-2 variance in the model indicates that the usefulness of ICC values as

well as the necessity of multi-level analysis are not valid. Thus, we interpret the rest of the result based on the general standard of binary logistic analysis.

First, the results showed that females were more likely to feel insecure than males. The value of odds ratio indicates that female respondents are three times more likely to express FoC than males. This observation is consistent with the result of previous studies [35,39]. This result also suggests that vulnerable age groups (minors and the elderly) tend to express higher level of FoC than middle-aged adults (20 to 49 years old). The odds ratio indicates that the likelihood of expressing FoC is almost five times higher for minors and the elderly than for adults. This result confirms again the theory of physical vulnerability, which states that people who are more physically vulnerable (minors, the elderly, and female) are likely to exhibit a higher level of FoC.

Second, the models revealed that people with a strong sense of community were more likely to feel insecure than people with a weak sense of community. This can be explained by the fact that there is an informal network of mutual support between strongly tied community members. It follows the same logic as Social Network Theory, which states that communities in which members are better connected are more e ffective collectively in reducing FoC.

Thirdly, the coe fficient of the 'income level' variable showed that the respondents with higher monthly income than the average (about \$117 per person) expressed higher level of FoC. This result contrasts with the principle of social vulnerability, which is based on the assumption that socially vulnerable people, particularly minorities and low-income people, express a higher level of FoC [52]. This can be explained by the local context that, as high-income people are in the minority, they must take a greater e ffort to secure their goods and properties. In fact, in the city, or even in the country as a whole, a high rate of poverty, unemployment, and crime overflows. The findings explained so far are true for both models with and without intersection-level variables.

With respect to the intersection-level variables, two density variables showed significant association with FoC at the 0.1 probability level; however, the signs conflicted with each other. This means that the relationships between pedestrian density of intersection and the expressed FoC seem to be convex curves with the minimum value. Meanwhile, the choice to include the two variables ("density" and "density squared") in our model comes from the fact that the results obtained by introducing each of these variables separately proved not to be significant. Indeed, when we introduce only the variable "density" in Model 3 and remove the variable "density squared", we obtain a coe fficient equal to 0.899 and a *p*-value equal to 0.257 for this variable. On the other hand, when we introduce only the variable "density squared" and remove the variable "density", we obtain a coe fficient equal to 0.501 and a *p*-value equal to 0.196 for this variable. Although the *p*-values are not statistically significant at a 0.1 probability level, both variables commonly showed positive signs, in contrast to the results in Model 3. This means that when we control for one of them, the other's e ffects can be more accurately identified. This also reveals that the quadratic regression approach of this study is more e ffective than the general approach using single density variable to identify the non-linear relationships between the key variables. Thus, we can conclude that Jacob's concept of 'eyes on the street' is not valid in places where pedestrian density exceeds a certain threshold level like Yaoundé city intersections considered in this study. We discuss the implication of this result in the following section.
