Replicating Ndii: Productivity Between Counties and Regions
David Ndii recently raised an important issue regarding how we measure the wealth of counties and whether we can accurately deduce a region’s wealth based on its contribution to GDP and other measures. David Ndii also went ahead to debunk what appears to be a fascinating myth that Nairobi is the wealthiest or most productive county. To back up his assertions, Ndii linked to a 2018 article that he published in The Elephant titled “The Politics of County Economies: Why Central Kenya MPs Are Wrong.” From the article, Ndii also questions the validity of the claim that Central Kenya is the most productive or wealthiest county. In his opening paragraph, he raises the issue:
Can Central Kenya contribute 60 percent of Kenya’s GDP, as recently claimed, nay, asserted, by the region’s members of parliament? If 12 percent (old Central Province) or 20 percent (including Meru, Embu, Tharaka Nithi, and Laikipia) percent of the workforce is responsible for 60 percent of the economy, what does that say about the rest of the country? What would make Kikuyus or GEMA community four or five times more productive than other Kenyans?
Readers of this blog will quickly note that I have, on numerous occasions, presented data that argued Central Kenya might be wealthier than other regions. My conclusions were informed by surveys looking at poverty rates, wealth distribution, and even GDP per Capita figures. The latter is the least reliable, and I only use it sparingly; however, from poverty and wealth statistics, there is a consistency in the data. The questions that David Ndii raises, therefore, seek to challenge what we currently know about Central Kenya. His critique is warranted, especially considering the unreliability of GDP figures and his more salient observation that counties cannot contribute to the economy since they mutually exchange value, and it’s hard to say one county contributed more than the other. As he posits, “When Nyandarua grows potatoes that are consumed in Nairobi: which county has contributed more to the other’s economy? Neither — they have exchanged value, with each profiting from the other.”
In this post, I will test out the claims David Ndii made in his article to see whether they hold up as they should. A caveat is that I am neither an economist nor a person of Ndii’s caliber1, so what I say can only be taken with a pinch of salt. One claim that I will test is whether Central Kenya is wealthy as per Ndii’s criteria. His criterion debunks the story told by GDP figures by laying more emphasis on labor output and consumption expenditures. Furthermore, I will also use night light data, which Ndii alludes to, to test whether there’s some consistency in the wealth differences between counties.
First, Ndii posits that rather than relying on GDP figures, which face the pitfalls raised earlier, we can more reliably measure the productivity of the counties by looking at what a worker in each county contributes to the national GDP. The GDP per worker is computed from each county’s share of the labor force. This criterion is more reliable than using GDP figures alone since it tells us in plain terms how much each worker in a specific county contributes to the national GDP. It effectively filters out the noise that comes with GDP figures, such as the number of people in a county, capital investments, and so forth. Based on this method, Ndii’s analysis raises exciting findings.
For instance, Nairobi is not the strongest economy. Kiambu leads with a GDP per worker of Ksh. 673,000. Other counties in the top 10 are Kwale, Nyeri, Nakuru, Kajiado, Laikipia, Murang’a, Garissa, Kilifi, and Machakos. Nairobi ranks 14th with a GDP per worker of Ksh. 302,000 and Mombasa 15th at Ksh 300,000. These findings support Ndii’s claim that Nairobi has a weaker economy. However, by this metric, one would ask whether his claim about Central Kenya and GEMA holds up to his data. Worth noting is that Central and GEMA are words that denote regions and not counties. Therefore, when talking about Central, we are not talking about a specific county. Instead, it’s about the five Kikuyu counties that make up the previous Central Province. To non-Kenyan readers, GEMA is a political term referring to Mt. Kenya tribes (Gikuyu, Embu, Meru). It consists of counties like Meru, Embu, Tharaka Nithi, Laikipia, and Nakuru (Plus the 5 Central Kenya counties). Ndii’s assertion, therefore, needs to be tested from a regional level rather than a county level.
Using the data shared by Ndii himself, I plotted a dual Y-axis chart below to represent the mean GDP per Worker2 for each region on the left y-axis (Line graph) and Mean GDP per Capita3 on the right y-axis (Bar graph). These data could each be plotted on a different chart, but I joined them to make my analysis more intuitive and to make comparisons easier. I also rank ethnic regions using Ndii’s criteria of GDP per Worker rather than GDP per Capita — That’s why the line graph descends smoothly while the bar graph is more rugged). From the ranking, we can see which regions contribute more per worker to the GDP and by how much.
Contrary to Ndii’s assertion, Central Kenya doesn’t contribute less in per-worker terms than any of the other regions. The Kikuyu counties that constitute Central average around Ksh. 450,000 in GDP per Worker. The Coastal Counties are next at around Ksh 350,000, followed by an almost equal amount for Maasai land (Kajiado and Narok) and Mt. Kenya. Nairobi county is fifth at Ksh 302,000, followed by the Somali counties. So, what does this tell us? Even in per worker terms, against Ndii’s remarks, Central Kenya still leads the rest of the country in its contribution to the GDP4. One might argue that Kiambu is doing a lot of heavy lifting and significantly raises the Central Kenya mean. Without it, Central still leads with a mean per worker contribution to GDP of around Ksh. 380,000, slightly above Coast. Ndii should have noticed that in his top 10 wealthiest counties, 3 (Kiambu, Nyeri, Murang’a) are from Central, and two are from Mt. Kenya (Nakuru, Laikipia). Overall, five of the counties are from the Mt. Kenya region. How did he fail to see this? I’d argue he looked too much into Nyandarua County, which was the only county in Central Kenya contributing less than Ksh 200,000 per Worker to the overall GDP. Kirinyaga was 12th and did better than Nairobi and Mombasa.
Since Ndii has dissuaded us from using GDP figures, let’s check how it compares with what we’ve found based on the GDP per Worker figures. Essentially, if the latter offers a more accurate picture of county economies, then how much do GDP figures overestimate or underestimate the contribution of counties? One interesting finding is that the correlation between these variables is positive but insignificant and weak (r=0.1, p=0.2). This should tell us that these two variables do not tell us the same thing, and we should expect massive discrepancies between them.
For example, we notice that GDP per Capita (represented by bars) underestimates the economies of some counties when compared to GDP per Worker. We know this by looking at the line and the bar for each region. If the line passes above the graph, then GDP figures underestimate the economy of that region. If the bar cuts through the line, as we see with Nairobi, then GDP figures overestimate the strength of that county’s economy. The distance between the two points tells us whether the error is huge or small.
What we see is that even though Central Kenya has a GDP per Capita higher than all other regions, GDP underestimates the strength of its economies compared to the GDP per worker. The same is true of the coast region, Maasai Land, Mt. Kenya, Somali region, Ukambani, Luhya Land, Luo Land, Turkana, and Samburu. Some regions are highly underestimated compared to others. Examples are the Somali region, Maasai, Coast, Kikuyu, and Mt. Kenya. Other regions do not have a significant difference between their GDP figures and their per-worker figures. These are Ukambani, Luhya, Luo, Turkana, and Samburu.
As Ndii observed, GDP figures overestimate the economies of some counties such as Nairobi, Kalenjin Land, and Kisii Land. It’s worth looking at what could be the reason for this, especially for the last two countries5. As for Nairobi, Ndii argues that there’s a huge difference between its GDP and household expenditure, with Nairobians spending around 40% on housing costs, especially renting.
Regional Disparity in Economic Growth
Even though David Ndii is skeptical of the economic dominance of Central Kenya counties, he appears to acknowledge that some regions are wealthier than others. He posits that:
At the other (lower) end of the scale, we have Elgeyo Marakwet and Nyamira with a GDP share, which is 40% of the labor force share and 12 counties where it is 50 percent. Looking at this pattern, it is readily apparent that the counties at the top are generally wealthier, while those at the bottom are poorer. The wealthier counties have more capital.
His remarks shouldn’t be surprising since even from a regional perspective, as portrayed by the chart above, there is a considerable difference across regions, with counties from Luhya land onwards being impoverished. He attributes the differences in wealth between the areas to capital investments, another variable that needs to be looked at separately to ascertain its validity. For example, one might ask, does Nyeri have more capital than other counties, considering it’s the third wealthiest county per Ndii’s criteria? This will need more research and a different article.
Currently, I’ll use another of Ndii’s criteria, which relies on night light data to estimate the wealth of a region6. The method relies on satellite imagery to map out regions with more lights at night. These regions are assumed to be more developed than their counterparts. For my analysis, I exclude Nairobi, Kiambu, and Nakuru for their extreme and disproportionate amount of night light, which makes them outliers. Nairobi has a sum of 17,754 night-lights, Kiambu 14,585, and Nakuru 10,466. All other counties have less than 5,000 lights, even though disparities might also exist between counties such as Kajiado, with 4,628 lights, and Narok, with 335. The bar chart below shows how different ethnic regions differ in terms of lights at night.
Even after excluding Kiambu, Kikuyu counties still have more night light than the rest of the counties. Specifically, Nyeri is the only rural county with a very high number of night lights, with 4,617 lights, more than Mombasa 4,597, Machakos 4,404, and slightly less than Kajiado 4,628, which is more urban when you consider places like Ngong and Kiserian. The idea that Nyeri is an old sleeping town isn’t based on fact. Many other rural areas, especially in Western and Nyanza, don’t have a lot of night lights, partly because they are primarily rural and because they are predominantly agricultural regions. Do night lights, therefore, tell us about the economy of an area?
First, there is a positive and significant correlation7 (r=0.6, p=0.000) between night lights and GDP per Worker that Ndii strongly emphasized was an accurate measure of a county’s economy. Night lights, therefore, tell us the same story that labor participation figures told us before, albeit not perfectly. The correlation with GDP per Capita is slightly lower yet significant and positive (r=0.48, p=0.001). Night lights also correlate positively and significantly to county HDI (r=0.52, p=0.000). There were negative correlations between overall poverty rates, night lights, HDI, GDP per Capita, and GDP per Worker.
Based on these findings, it’s safe to say that different regions have different levels of development and that some regions contribute more to the Kenyan economy than others. Contrary to Ndii’s assertion, this analysis has found that Central Kenya still contributes more to the national economy than most other regions as measured by labor force participation and GDP contribution per worker in each county. From Ndii’s analysis, 3 out of 10 wealthiest counties are from Central Kenya, while a total of 5 out of 10 constitute GEMA or Mt. Kenya counties. Ndii was correct to observe that some countries contribute less to the economy despite their high GDPs. From my analysis, this was true of Nairobi, Kisii Land, and Kalenjin Land. Finally, data from night lights corroborates these findings, revealing that some regions are likely to be more developed than others. However, this method underestimates agricultural wealth from most counties, especially those in rural areas. Only Nyeri stands out as a rural area with a significant number of night lights. Overall, Central has a larger share of nigh lights than any other region in the county after excluding Kiambu county, a finding consistent with both GDP per Capita and GDP per Worker figures.
1 David Ndii is the Chief Economic Advisor to President Ruto.
2 Ndii shared the Figures for GDP Per Worker through his article in The Elephant and also through his X handle here. These figures are from 2016, which is more than half a decade ago, and a lot might have changed since then. It’s, therefore, essential to countercheck with more recent data.
3 These GDP Per Capita figures are more recent from 2019. Using them on a dual Y-axis chart, together with GDP Per Worker data, might conceal the fact that the time scales are different and do not represent economic development during the same period. 2016 figures might have been more appropriate but are unavailable since KNBS only started collecting county-level GDP figures in 2019.
4 This post does not attempt to quantify by how much Central Kenya contributes to the economy. It only shows that these counties indeed contribute more than others when aggregated.
5 These two regions need further investigation.
6 Ndii gets his information on night lights from an article by the World Bank titled Bright Lights, Big Cities Measuring National and Subnational Economic Growth in Africa from Outer Space, with an Application to Kenya and Rwanda. You can access the paper here. I use night light data from this article to create the chart above. These data come from 2013, a decade ago. Recent data would be more appropriate.
7 Correlation coefficients range from -1 to 1 with 0 between. -1 means a perfect negative correlation, while +1 indicates a perfect positive correlation. Values close to 1 are said to be high correlations, 0.5 is medium, and 1–4 is low correlation. The same applies to negative correlations. 0 means no correlation at all.
