factors.Rmd

---
title: "Factors that Influence the Activity Level of Time Banks Associated with hOurworld"
output:
  html_document:
    theme: united
---
### Theoretical Framework
A number of quantitative factors would in theory affect the activity level of a timebank initiative, namely: the size of its membership size, the ratio of active members, and the number of admin accounts.

First, a large membership in theory would encourage more active exchanges among time bank members due to the higher number of exchange requests being made as well as the higher diversity of skills and labours offered. Ideally, this would promote vibrant time credit exchange among members, thereby increases the staying power of the initiative. And with the help of the software platform, a larger membership should not have a positive correlation with the labor of time bank coordinators.

Second, a high ratio of active membership participation would in theory lead to a more vibrant time bank. This is because a low membership participation rate can seriously jeopardize the matchmaking of the time bank software system, undermining the user experience of time bank members while significantly increasing the workload of coordinators. This is because when a time bank member makes an exchange request, the software system will automatically generate a short list of individuals who offer relevant skills on their user profile. This gives the requesting individual a guideline of who he or she should contact with and seek help from. However, when the percentage of active members is low, the majority of the options offered on that list will not be reachable. This would be extremely frustrating for the individual who make the request.

Last but not least, time banks with more admin accounts would in theory mean that there are more time bank coordinators participating in the management and advocacy of the initiative, and is therefore a good indicator of a higher institutional capacity. This means that the coordinator team would be more effective and responsive in addressing the needs and concerns of time bank members. This in return would improve the user experience of time bank members, giving them more incentive in participating time credit exchanges.

### Data visualizing and modelling
By using the data provided by hOUrworld, this project tries to see if the correlation between the three factors and the operational vibrancy of time banks coincide with our theoretical deduction. The activity level of a time bank is measured by the hour exchange per membership (exchange per capita) in the past year (365 days). Because many time banks are startups with minimal exchange, this analysis only focuses on time banks that are launched before 2016 and at least have 10 hrs of exchange in history. The result is as following:

![Fig.1 Correlation between membership size and exchange per capita of time banks associated with hOurworld in the past year](graphics/scatter_activity-membership.png)

![Fig.2 Correlation between active membership ratio and exchange per capita of time banks associated with hOurworld in the past year](graphics/scatter_activity-ratio.png)

![Fig.3 Correlation between active membership ratio and exchange per capita of time banks associated with hOurworld in the past year](graphics/scatter_activity-admin.png)

Based on these diagrams, it is clearly evident that there is not a visible or statistically significant correlation between the activity level of time banks with any of the three variables we have identified. The outliers are both numerous and extreme.

Given the obvious inadequecy of regression model, this project also incorporates a decision tree to understand how these three factors may influence the performance time banks from a more detailed perspective. Any time bank with an exchange per capita lower or equals to 2 hours in the past year is identified as an inactive time bank. Those with an exchange rate higher than 2 hours per capita is identified as an active time bank. Based on this premise, the decision tree is as following:

![Fig.4 Decision tree of time bank activity level](graphics/hOurworld_tree.png)

Based on the decision map, it seems that time banks with a higher number of administrator accounts are almost always favoured over time banks with a smaller number of administrator accounts in having an active institution status. The ratio of active membership has a minimal influence over whether a time bank initiative is active or not is minimal. The impact of the membership size, however, is more complicated. As a rule of thumb, small time banks with a membership of 48 or less tend to be inactive, while the active level of larger time banks is often partially determined by the number of administrator accounts based on this decision tree model. Based on random forest test, this decision tree model has a **class error** of **0.432** for active time banks and a **class error** of **0.35** for inactive time banks. Therefore, caution need to be taken when using this model as a tool for prediction.

The mean decrease gini of each variable is as following:

![Fig.5](graphics/Mean_Decrease_Gini.png)

The mean decrease gini of membership is high. The mean decrease gini of the number of admin accounts is moderately high. The mean decrease gini of the active membership ratio is very low.

### Data Interpretation
#### Administration Account
The number of adminstration accounts of the time bank is a variable with relatively high mean decrease gini in the decision tree model, and having a higher number of administration account always leads to a prediction of an active time bank. This result conicides with our theoretical hypothesis to a certain extent, where a larger team of time bank coordinators would encourage a more active time bank. However, due to the low threshold for an "active time bank" used in this decision tree model is very low - a meagre exchange rate higher than 2 hours per capita is required to satisfy this category. This, coupled with the fact that the truly vibrant time banks with a high exchange rate of more than 10 hours per capita are a rarity in this dataset, means that the correlation between the number of administration accounts and the activity level of a time bank remain inconclusive.

The scatter plot of the number of coordinators against the exchange per capita for the past year does not indicate a clear linear correlation between these two variables. This might be caused by the fact that some small time banks with small membership - and therefore requires a tiny team of coordinators - have a high extrange rate per capita. Many time banks rely on volunteers for their daily operation, and pay these volunteering members by time credits. For small time banks, these exchanges can represent a significant proportion of their total exchange volune. The exchange among members, on the other hand, may be rather limited. But due to the small operational scale, the exchange per capita of some smaller time banks could be very high. This obscures the correlational pattern of the scatter plot.

Last but not least, one should be noted that the number of administration account may not be as good of an indicator of the institutional capacity of a time bank. This is because the time commitment and professional skills between different coordinators may vary. For instance, a full-time paid coordinator would have a significantly higher contribution to the institutional capacity of a time bank network than a volunteer coordinator who only oversees the operation of the initiative occationally. Therefore, to outline correlation between institutional capacity and membership participation of time banks, more detailed analysis is needed.

#### Membership Size
The scatterplot does not provide any identifiable correlation that is statistically significant. There seems to be a loose positive correlation among time banks with a membership smaller than 500, but due to numerous outlying values the correlation is rather weak. This is likely to be the result of small time banks with high exchange rate per capita similar to the scenario seen in the case of administration accounts. In addition, this might also be caused by the presence of dormant time banks with a relatively large membership in the dataset. Dormant time banks are those time banks that had experienced some level of success in their operation history, but have now lost their vibrancy. These time banks often tend to have a relatively large membership size, but their exchange in the recent years could be minimal.

Membership size has a high mean decrease gini in the decision tree model, meaning that it is a relatively reliable indicator in predictions. However, the correlation is not as straight forward as the correlation between membership size and time bank activity. For time banks with a small number of administration accounts, a larger membership size often leads to predictions that favour an inactive time bank. This is a really interesting finding, and to some extent proves the importance of institutional capacity for time banks to successfully operate. One of the mose important reasons for time banks to go dormant is the burning out of coordinators, where time bank coordinators are overwhlemed by their obligation and quit, leaving an orphaned time bank. While a small team of coordinators may be sufficient in overseeing the daily operation of a time bank with small membership, they are more likely to burn out when they are managing a large time bank. Hour decision tree model supports this theory.

#### Membership Participation Rate
The correlation between membership participation rate and the exchange per capita for the past year is minimal in both the scatter plot and the decision tree model. However, this does not mean that it is not a relevant factor. Rather, this result is likely due to the poor quality of data in this data set. Using the time bank catalog, it is evident thatmany time banks with minimal exchange for the past year reports a membership participation rate of 100%. This corrupts the dataset, making this variable devoid of any analytic value. More rigorous data collecting is required. 

### Conclusion and Future Prospect
The result of this quantitative analysis of the three factors is rather underwhelming, but it has nontheless provide us a path for future research. First, we need to provide a measure for time bank activity that balance the exchange per capita for the past year and the total exchange volume. Through this analysis, it is evident that using the exchange per capita for the past year as a measure for the activity level of time banks favours small time banks. But the frequency of exchange is only part of the story. Time banks with a large exchange volume have a greater impact to their community as a whole, and the socioeconomic influence they could generate should also be recognized.

Second, the quality of the data should be reviewed. The irregularity we have seen in this analysis with the active membership ratio is clearly due to the low quality of teh data, and this should be addressed as quickly as possible.

Last but not least, this research highlights the importance of supplementing the quantitative research on time banks with qualitative methods, which would probably give us a richer, more dynamic model for future analysis.

Xu, Haitong 2016