A Geographic Overview of Bangladesh

## OGR data source with driver: ESRI Shapefile 
## Source: "shp", layer: "bgd_polbnda_adm2_wfp"
## with 64 features
## It has 4 fields

The figure above shows Average Revenue, denominated in local currency (Takas), in the year 2009 for businesses in the surveyed districts, A few things are made apparent on this plot. First of all, the survey was conducted only in a selection of districts, not the whole country, and only the surveyed districts are filled with some kind of color.

Are there any clear geographic patterns to annual revenue? Such a conclusion is difficult to draw based on the information on this map. The district with the lowest average revenue per firm (shaded in darkrest blue) appears to be right in the center of the country. There are a number of other districts that have comparable average revenue. The district with the highest average revenue is shaded in light blue inn the southwest of the country; no other district seems to have a comparably high average revenue per firm.

Now, let’s take a closer look at the data for annual revenue.

Analysis of Annual Revenue variable

Let’s explore the Annual Revenue variable through a series of histograms.

Above histograms display distribution of values of annual revenue for 2009. I chose to use thre histograms to display the same variable, with different binwidths, which allows us to view the data from an increasingly generalized perspective. We can see that the distribution is not normal–it is a negative exponential distribution (skewed to the left). The distribution makes sense, if one expects that most firms in Bangladesh are small and would not have a high revenue.

What does the differencces in binwidth tell us? The histogram with binwidth of 100000 demonstrates that the data periodically spikes, and does not appear as smooth as the next 2 histogramms. These spikes likely result because survey respondents would tend to give estimated answers, corresponding to roundded increments.

The second histogram reduces the number of spikes substantially, and the third almost eliminates them entirely, creating a smooth staircase shape. I believe that the second histogram sufficiently manages the spikes without totally eliminating the variety of responses, so I believe that one is the best to use.

Applying logarithmic adjustment, we can create a new histogram:

The logarithmic adjustment of the annual revenue variable creates a new histogram that displays the data as a normal distribution.

Let’s take another view of business size, by comparing annual revenue to number of employees.

The plot above displays number of employees in the firm on the x axis, and annual revenue on the y axis. It is difficult to discern any clear pattern; We appear to have a problem with overplotting, as many of the data points appear to overlap. This probably results from the fact that number of emmployes is an integer, and is limited on the graph to a value of 20. Let’s add a jitter to the plot to make things clearer.

The jitter does help somewhat to make clear the varied relationship between number of employees and annual revenue. It makes sense that there is an overall positive relationship, but based on this plot, the relationship does not appear to be that strong. There seems to be many instances of a low number of employees with high revenue, and high number employees with a low revenue.

Let’s take another view of this relationship, by applying a logarithmic scale to the annual revenue, and maintaining the jitter.

The application of this change seems to make clearer a positive relationship between annual revenue and number of employees.

Let’s numerically verify this relationship by calculating the correlation between number of employees and annual revenue.

## 
##  Pearson's product-moment correlation
## 
## data:  surveydata$q4_7 and surveydata$q1_12a5a
## t = 14.8642, df = 1722, p-value < 2.2e-16
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  0.2947005 0.3784047
## sample estimates:
##       cor 
## 0.3372189

The correlation value of .337 indicates that there is a relationship between the number of employees and annual revenue, albeit it cannot be chacterised as a strong relationship.

Analysis of Business Classifications

What type of businesses were surveyed, and is it possible to explore some of their characteristics? Perhaps a more insightful question would be, is it possible to determine if annual revenue varies by business sector?

The survey data includes information about the industry classification of the business; the business information is given via the ISIC code, an industry classification system. ISIC is a 4-tiered classification system, with the lower 3 tiers represented by a numeric code, and a corresponding top level represented by a letter. The lowest (and most granular) level is represented by a 4-digit code.

The following 2 lists and pie chart are intended to explore through the different classification levels within the data. I wish to find the right level of granularity to the classification data that can further facilitate analysis. There is a balance to be achieved; granular classifications could yield granular insights, but may require an unwieldy number of posible classifications; too few classifications could inhibit any meaningful analysis, because we need a critical mass of observations to make good ,generalizable observations for classifications of data.

The following code imports the ISIC code descriptions, and adds the descriptions to the survey data. The ISIC classification descriptions were sourced from the “[UN Department of Economic and Social Affairs] (http://unstats.un.org/unsd/cr/registry/regcst.asp?Cl=27)”; some manipulation of this data was required to shorten the category descriptions, for better display of names on the plots, and to create better mapping between different codes.

## Source: local data frame [4 x 2]
## 
##          ISIC2_desc    n
## 1                   1699
## 2     Communication   18
## 3    Land Transport    6
## 4 Transport Support    1

## Source: local data frame [76 x 2]
## 
##                                                                       ISIC4_desc   n
## 1                  Retail sale of textiles, clothing, footwear and leather goods 217
## 2                                                 Restaurants, bars and canteens 179
## 3                                    Other retail sale in non-specialized stores  99
## 4                                       Manufacture of structural metal products  92
## 5                                        Other retail sale in specialized stores  86
## 6                    Retail sale of household appliances, articles and equipment  83
## 7                             Manufacture of wearing apparel, except fur apparel  82
## 8  Retail sale of pharmaceutical and medical goods, cosmetic and toilet articles  68
## 9                                                       Manufacture of furniture  64
## 10              Retail sale of food, beverages and tobacco in specialized stores  58
## ..                                                                           ... ...

## surveydata$ISIC_sectiondesc: Health and social work
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##   80000   96000  112000  112000  128000  144000 
## --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 
## surveydata$ISIC_sectiondesc: Hotels and restaurants
##     Min.  1st Qu.   Median     Mean  3rd Qu.     Max. 
##    90000  1400000  2948000  4073000  5010000 28000000 
## --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 
## surveydata$ISIC_sectiondesc: Logistics & Communications
##     Min.  1st Qu.   Median     Mean  3rd Qu.     Max. 
##   200000  1300000  2020000  5711000  7000000 44400000 
## --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 
## surveydata$ISIC_sectiondesc: Manufacturing
##       Min.    1st Qu.     Median       Mean    3rd Qu.       Max. 
##        999     700000    1680000   19300000    6500000 1860000000 
## --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 
## surveydata$ISIC_sectiondesc: Other community
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##   16000  190000  330000  670300  600000 9600000 
## --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 
## surveydata$ISIC_sectiondesc: Real estate
##     Min.  1st Qu.   Median     Mean  3rd Qu.     Max. 
##    60000   240000   635000  1283000  1350000 11600000 
## --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 
## surveydata$ISIC_sectiondesc: Retail & Repair
##      Min.   1st Qu.    Median      Mean   3rd Qu.      Max. 
##       999    850000   2184000   9384000   5400000 600000000

Exploring Business Size- Revenue vs Number of Employees

As another measure of business size, how does the number of employees vary across ISIC sections?

Based on this boxplot, section Hotels & Restaurants has the highest median number of employees, followed by Manufacturing. The medians for those sections are significantly higher than the other sections, and the difference betweeen 1st and 3rd quartiles wider. ###Point Plot comparison - Annual Revenue & Number of Employees by ISIC section Now, let’s explore the two size characteristics together with the ISIC section, all on the same point plot.

And above plot, but applying logarithmic scale to annual revenue:

What is apparent in the first plot is made more explicit in the second plot (the one with the logarithmic adjustment applied to annual revenue): how intensively different ISIC sections utilise labor. ISIC section Retail & Repair has revenue that grows very quickly for a small growth of number of employees; however, in sectors Hotels & Restaurants and Manufacturing, annual revenue increases more slowly as number of employees grow–implying that those sectors are more dependent on labor for growth.

## 
##  Pearson's product-moment correlation
## 
## data:  q4_7 and q1_12a5a
## t = 8.1324, df = 479, p-value = 0.000000000000003553
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  0.2672266 0.4244997
## sample estimates:
##      cor 
## 0.348312

## 
##  Pearson's product-moment correlation
## 
## data:  q4_7 and q1_12a5a
## t = 12.2992, df = 907, p-value < 0.00000000000000022
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  0.3209411 0.4324666
## sample estimates:
##       cor 
## 0.3780747

## 
##  Pearson's product-moment correlation
## 
## data:  q4_7 and q1_12a5a
## t = 19.7177, df = 178, p-value < 0.00000000000000022
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  0.7759546 0.8691912
## sample estimates:
##       cor 
## 0.8282201

## 
##  Pearson's product-moment correlation
## 
## data:  q4_7 and q1_12a5a
## t = 1.7646, df = 63, p-value = 0.08247
## alternative hypothesis: true correlation is not equal to 0
## 95 percent confidence interval:
##  -0.02837827  0.43775115
## sample estimates:
##       cor 
## 0.2170229

Exploring the sources of financing for businesses

Now, let’s explore some of the other characteristics of the business. The dataset contains information about sources of loans. The respondent is asked to give a series of yes / no answers about whether the business takes credit from various types of sources, such as banks or moneylenders. In order to create a plot that can allow us to compare the responses, the data requires some manipulation to achieve the right input format. I used the “melt” function in the “reshape2” function to do so.

These faceted boxplot reveal that the sources of credit (x axis, 1= Yes, 2 = No) varied widely by annual revenue (y axis). Government and Private Banks showed by far a higher median revenue than the other options, and the difference between 1st and 3rd quartiles is mmuch wider than the other sources, indicating that the Banks lent a wider variety of larger amounts. Firms with smaller revenue turned to Family and Friends, Moneylenders, and Microfinance organisations for credit; that they would do so is not surprising, considering that banks have traditionally lent to larger and more established firms. Microfinance organisations, moneylenders, and family & friends are often sources of credit for busineses who have trouble obtaining loans from banks.

## surveyfinance_yes$financesource: Family and Friends
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##    10.0   354.0   698.0   795.7  1263.0  1684.0 
## --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 
## surveyfinance_yes$financesource: Government bank
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##    11.0   687.8   967.0   931.7  1211.0  1725.0 
## --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 
## surveyfinance_yes$financesource: Microfinance Organization
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##     2.0   476.0   954.0   927.2  1428.0  1717.0 
## --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 
## surveyfinance_yes$financesource: Moneylender
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##    50.0   572.8  1006.0   915.2  1269.0  1725.0 
## --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 
## surveyfinance_yes$financesource: Other
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##      42      83    1352     855    1396    1509 
## --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 
## surveyfinance_yes$financesource: Private bank
##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##    10.0   415.5   855.0   854.2  1272.0  1722.0

A look at the actual summary statistics for the loan data confirms above assessment: median loan size for Government Banks (6350000) and Private Banks (4800000) are far higher than the other sources of credit; the closest is Moneylenders at 23000000 takas.

Does the lending vary by sector?

Let’s look at how the different categories of lenders lend in different sectors.

Plot One

Description One

My first final plot is the map of Bangladesh that we used near the beginning of this project, adding some features to provide some context. First of all, I plotted on the map the 3 largest cities in Bangladesh, with orange circles indicating population size of each. As wealth often concentrates in urban areas (typicallly being centers of economic activity), I wanted to test the idea that firms with the largest level of revenue coincided with major urban centers in Bangladesh. In fact, in this case, we note that the highest average revenue coincides with the three major urban centers; Khulna district has the highest average revenue of 29 million takas, and also contains the city of Khulna, the 3rd largest in Bangladesh. Chittagong district, with the second highest average revenue (15.5 million takas), also contains the second-largest city in Bangladesh. And the district with the 3rd highest average revenue (15.1 million takas) is Dhaka district, where the capital and largest city of Bangladesh is located.

So, it seems that urbanization in Bangladesh coincides with high annual revenue for firms in those urban locations.

Plot Two

Description Two

Plot two builds on the point plot comparison above, and refines to add trendlines to the plot. This plot has the following features:

• A log10 scale is applied to the Y axis (Annual Revenue 2009);
• Jitter is applied to avoid overclustering of points;
• The points are colored by ISIC section;
• Trendlines are added for select ISIC sections, to highlight the differences between them.

The trendlines emphasize the findings from the original plot: within the Retail & Repair section, annual revenue increases at a distinctly faster rate as employees increase, than in the Manufacturing andd Hotels and restaurants section. Using the combination of plotting methods, we have been able to identify a differentiating characteristic between certain sectors: at higher levels of revenue, Retail & Repair firms feature fewer employees than the Manufacturing and Hotels & Restaurants sections.

Plot Three

Description Three

Plot three is a refinement of the box plots that were explored above. Those box plots informed the creation of plot three in an number of ways.

Firrst, to make an effective and attractive visualization that captures the salient observations of our findings, it would be an idea to limit the number of items within the categories on the plot. For the ISIC section variable, I selected items that have enough observations to be able to draw meaningful insights; we can tell from the box plots and the point plots that items within the ISIC section category, such as “Health & Social Work”" and “Other Community” do not have many observations–so it is difficult to glean any insights about those sections. I selected 3 sections: “Hotels and restaurants”, “Manufacturing”, and “Retail and Repair”–the items that I also highlighted in plot two with trend lines. Also, considering the other categorical variable, Source of Financing, I had noted that the 3 items from ISIC Section had displayed some distinctive characteristics as the source of financing change; additionally, I also included “Logistics and Communications”, as for the Source of Financing “Microfinance Organization”, the revenue patterns were distinctive when compared to the other ISIC sections.

Second, I noted that when faceting the Source of Financing “Government Banks”, it was difficult to create a scale that could adequately show the range of responses for annual revenue; scaling on the y axis that appropriately showed the “Manufacturing” seciton would distort the plots of the other sections. To resolve this issue, I applied the “coord_flip” option to make a horizontal boxplot.

Third, I saw that being specific about the order of the data would help to bring out the story of the data. So, I adjusted the input data frame to order both categorical variables, to generally place the ISIC sections and sources of financing with higher annual revenue towards the top of the plot, while maintaining the order of sources of financing within each sub-plot to facilitate comparison.

This plot captures the relevant information that we gathered from the box plots. We can observe that government banks tend to provide credit to firms with higher levels of revenue, while moneylenders and microfinance organizations tend to lend to firms with relatively lower levels of revenue. We can also see the differentiated levels of revenue between the different ISIC sections, with Manufacturing firms having a markedly higher revenue than the other sectors.

Most interestingly, we can observe a potential segmentation of the finance market, by discerning patters of how different lenders tend to lend to different industries. Microfinance organizations are designed to lend to firms of a very small size–and so one would expect that the firms who receive credit from such sources would have a lower annual revenue. This seems to hold true in most of the sections; however, we can see that firms in Logistics & communications that receive microfinance loans have a median revenue that is much higher than microfinance loan recipients in other sections. Government banks, on the other hand, may tend to target credit towards larger firms–hence the phenomenon of firms in the Manufacturing sector that take government bank loans, that show much higher levels of revenue than in other sections; hence, manufactuers seem to be a good market segment for government banks. Indeed, the growth of manufacturing in Bangladesh has been a well-known story of success, so much so that it is probably considered to be a strategic industry in the country–something that governments would tend to encourage (see http://www.mckinsey.de/sites/mck_files/files/2011_McKinsey_Bangladesh.pdf for more information).