Presentation on theme: "Today’s schedule Finish “Democracy and Development” [Luo et al.] Summarize the use of IV’s Fixed Effects –Is Sharecropping Inefficient? [Shaban] p. 39."— Presentation transcript:
Today’s schedule Finish “Democracy and Development” [Luo et al.] Summarize the use of IV’s Fixed Effects –Is Sharecropping Inefficient? [Shaban] p. 39 –A simple example: Does irrigation affect yields in China’s agriculture? –Does Privatization lead to better performance in China’s rural industries? [Li and Rozelle] Start on: PSM / D-in-D / Experiments
Summarize the use of IV’s –Overcome simultaneity bias / measurement error / unobserved heterogeneity –Definitions of an IV Correlated with endogenous variable Affects outcome variable ONLY through its effect on the endogenous variable –Test for a good IV Statistical tests (necessary/not sufficient) – 5% Logical tests – 95%
Things to watch out for in your IV analysis a.) pass logic test … but first stage correlation is weak “weak IV” can mean second stage is measured insignificant / but, really significant b.) careful of “nature of IV” … what “part of variability” of the endogenous variable are you using c.) be careful of “old rules” and “simple fixes” Y = a0 + a1*X + a2Z + e … but worried about X being endog. Use: X t-1 as an IV for X t … lagging is better but far from enough All policy variables are good IVs in equations of private behavior … NO WAY … policy makers are not stupid, blind and irrational If you use fixed effects, don’t need IV … and vice versa … NOT necessarily!!!
Lecture 03 Fixed Effects, Sharecropping and the Privatization of China’s Rural Industries Scott Rozelle Stanford University
The Puzzle of Sharecropping Alfred Marshall lobbied Congress to have it banned in the US The Government of India still has many laws and regulations banning sharecropping Why? The argument is that “Sharecropping is inefficient” Yet: Steven Cheung, a development economist from Chicago and one of the earliest economists to try to “put institutions into the study of development” … argued that he can show that “sharecropping is NOT inefficient”
Introduction I choose this topic because sharecropping is an ancient institution that is prevalent in many parts of the world irrespective of whether or not it is “inefficient” It is an institution that involves a lot of economics (trade off between efficiency and risk, adverse selection, moral hazard, etc.). It has fascinated and continue to fascinate many development economists. There is a large literature on sharecropping tenancy. Indeed, some trace the first attempts by development economists to explain institutions and understand their role in the process of development to the study of sharecropping (Johnson, 1950s; Cheung, 1960s; etc.) It is a good way to teach “fixed effects”
Plan for Today What is Sharecropping? What is the (theoretical) “problem” with sharecropping? [why did Marshall hate it?] Is it really inefficient? [empirical approach to testing for the inefficiency of SC … answer is “yes”] So why do people choose it?
Jumping to the last point (because I will probably not have time to cover completely … really should take 1 or 2 classes to cover properly) Perhaps the best answer in the literature is: –Sharecropping (SC) is sometimes optimal in a second best world –Second best happens because “risk is important” … and “there are no credit or insurance markets in agriculture” [remember Binswanger and Rosenzweig, lecture 01) –This already helps explain: Why SC might be more prevalent in India (where the monsoons are so risky) … sunspots determine if the monsoon is big or small (and, a big or small monsoon determines if the harvest is big or small) … Why SC is higher when farm prices are low and interest rates are high (the risk to farmer of being able to cover his/her fixed cost rise … if there is a bad harvests and revenues are down, farmers can’t make interest payments, they can lose their farm … as many did in the 1980s)
What is Sharecropping? Sharecropping is a system of agricultural production where a landowner allows a sharecropper to use the land in return for a share of the crop produced on the land.agricultural production There are a wide range of different situations and types of agreement. The share varies from country to country also across regions within a country, although 50-50 division is more common. Input costs are sometime also shared between landlord and tenant.
Alternative Tenancy Contracts Fixed Wage: Landlord brings the land (her own) and pays the agricultural laborer a fixed wage Fixed Rent: Tenant brings the labor (his own family) and pays the landlord a fixed rent [in the literature, always comparing the efficiency of these alternative institutions …
The “Problem” of Sharecropping The institution of sharecropping has always had a bad name. ﹣ Alfred Marshall lobbied Congress to have it banned in the US ﹣ The Government of India still has many laws and regulations banning sharecropping Why does SC have such a bad reputation? -The argument is that “Sharecropping is inefficient” Because they get only part of the return, they will naturally reduce their effort … (sometimes called “Marshallian” inefficiency)
Why is it “Marshallian” inefficient? The problem from the view point of the tenant: Max Profits: Y s = *F(k s,n s ) + w(1-n s )(1) n s, Where Y s is the tenant’s income, is the share of the output that is taken by the tenant, F(.) is the output function, k s and n s are land and labor (normalized to 1) tenant puts on farm production, and w is the off farm wage, which produces off farm income of (1-n s )*w.
Optimizing Problem (1) Choosing capital and labor allocation, the first order conditions are:∂Y s /∂n s = *F n ’(.) - w = 0(2) ∂Y s /∂k s = *F k ’(.) = 0(3) The inefficiency is because F n ’(.)=w; the tenant under use labor (n) because he/she does not get full return to his/her effort. So would tenant put on more or less labor (n)? –Because the return is lower (only get of output, F) … to get equation (2) to balance … need to raise the value of F n ’(.) … this can happen if the labor (n) is reduced (because when there is diminishing marginal returns and F n ’’(.) < 0 … then F n ’(.) is larger when the amount of labor (n) is smaller ….
Optimization (2) What about land (k)? How much land would the tenant use if the contract did not have any contingencies against its use? ∂Y s /∂k s = *F k ’(.) = 0(3) Since F k ’(.) = 0, the tenant will use the land as much as possible. [this was first pointed out by D. Gale Johnson in 1950s, who then showed that it was because of this that sharecropping contracts typically had very detailed arrangements that told the tenant how he/she can use the land – e.g., like the date before which tractors can not be used].
But is sharecropping completely inefficiently? How do tenants use fertilizer? In many places in the world (including India and California) there is a sharing rule with fertilizer. For example, if the output is shared 50-50, then fertilizer is almost always shared 50-50. How does this help? Now tenant’s problem is to maximize Profits: Y s = *F(k s, n s, f s ) + w(1-n s ) - *p f * f s (4) n s, k s, f s Where f s is the amount of fertilizer put on the crop, p f is price of fertilizer, and the sharecropper only has to pay for of the fertilizer, the rest, 1- , being paid for by the landlord. In this case, the first order conditions are: ∂Y s / ∂n s = * F n - w = 0 (5) ∂Y s / ∂k s = * F k = 0 (6) ∂Y s / ∂f s = * F f - * p f = 0(7) and the “efficient” amount of fertilizer is being applied (F f = p f ).
Field-side Picnics on “Spreading fertilizer Day” ( for the landlord and his family ) What do you think happens on the day that fertilizer is being applied? Where is the landlord? If you guessed by the side of the field watching, that is right? If he was not watching, what might the tenant do? Of course: take the extra fertilizer to the market and sell it and put the money in his/her pocket … and “under-utilize” fertilizer (from the landlord’s point of view)
The inefficiency of sharecropping: A numerical example
Crop Production – on own land (or rented-in land for a fixed rent) Food Output -- kgs Weeks of Labor 12345678 3000 5400 5300 5100 4800 4400 3800 (Assume the tenant has only one variable input of production – labor) 5450
Economics of Farming on own land (5 acres) Each manweek is worth $300 to the household (could produce that much in garden in private plot … or that is how much leisure is valued) Profits from Farming own land Labor week Output (kgs) Marginal product (kg) Price for output Marginal increase in value ($) Marginal cost ($) 13000 $1/kg$3000$300 23800800 $1/kg$800$300 34400600 $1/kg$600$300 44800400 $1/kg$400$300 55100300 $1/kg$300 65300200 $1/kg$200$300 75400100 $1/kg$100$300 8545050 $1/kg$50$300 Point of profit maximization: marginal revenue equals to marginal cost … therefore, when household farming by itself, they will put in 5 weeks of labor and produce 5100 kgs of grain and earn: 3600 dollars
Farming and returns – by household itself Food Output -- kgs Weeks of Labor 12345678 3000 5400 5300 5100 4800 4400 3800
Crop Production – Sharecropping on 50-50 At end of season, split returns ( ½ and ½) with the landlord Food Output -- kgs Weeks of Tenant’s Labor 12345678 1500 2700 2650 2550 2400 2200 1900 Returns to landlord Returns to Tenants 1/2 4400
Each manweek is worth 300 RMB to the household (could produce that much in garden in private plot … or that is how much leisure is valued) Profits from Farming own land. Labor weeks Output (kgs) Marginal product (kg) Price for output Marginal increase in value ($) Marginal cost ($) Marginal product ( *Y) (kg) Marginal increase in value ($) 13000 $1/kg$3000$300 1500 $1500 23800800 $1/kg$800$300 400 $400 34400600 $1/kg$600$300 300 $300 44800400 $1/kg$400$300 200 $200 55100300 $1/kg$300 150 $150 65300200 $1/kg$200$300 100 $100 75400100 $1/kg$100$300 50 $50 8545050 $1/kg$50$300 0 $0 Sharecropping Economics of Farming on share tenancy: Note reduction in effort (from 5 weeks of work to 3 weeks … because only get part of return!
Farming and returns – Sharecropping on 50-50 basis At end of season, split returns ( ½ and ½) with your partner Food Output -- kgs Weeks of Labor 12345678 1500 2700 2650 2550 2400 2200 1900 Returns to landlord Returns to tenant 1/2
And sharecropping can be full efficient if… The “efficient” view of sharecropping is based on the work of D. Gale Johnson (1950) and especially Cheung (1968, 1969). The main argument is that worker’s effort can be monitored and enforced. Contracts offered by the landlord would stipulate in great detail regarding the size of plot, the tenant’s share, the intensity of cultivation, field and crop management, etc.
Farming and returns – Sharecropping on 50-50 basis At end of season, split returns ( ½ and ½) with your partner Food Output -- kgs Weeks of Labor 12345678 1500 2700 2650 2550 2400 2200 1900 Returns to landlord Returns to tenant 1/2 If labor is fully monitorable, landlord specifies that fixed wage worker or tenant put in 5 weeks
Summary of Progress So Far What is Sharecropping? What is the (theoretical) “problem” with sharecropping? [why did Marshall hate it?] Is it really inefficient? [empirical approach to testing for the inefficiency of SC … answer is “yes”] So why do people choose it? Now the question is:
Are Sharecropping really inefficient? Subject to empirical test So now we know what Marshallian inefficiency is … And, we know that it is theoretically a problem if work effort can’t perfectly be monitored and enforced. But the alternative is true (“efficient”) if it can be perfectly monitored and enforced. A BASIC QUESTION: is SC really inefficient, or which alternative argument is true? This need to be answered by empirical test.
First, test based on a “naïve” model Set up the “naïve” model Y = a 0 + a 1 *Z 1 + a 2 *Z 2 + a 3 *D sc + ε (2) Where Y is yield or Intensity of Input Use Z 1 are household/village characteristics--things such as: education of the farmer age of the farmer location of the village Z 2 are plot specific characteristics such as: quality of land other etc. D sc = a dummy variable … 1 if farmer is farming as Sharecropper and 0 if farmer is cultivating his own plot RESULTS: In line with the theory, a 3 <0 in most empirical studies. However, a 3 cannot be interpreted as the causal effect of incentives on productivity or input intensity, because…
Endogeneity.. True equation: Y = a 0 + a 1 *Z 1 + a 2 *Z 2 + μ n + a 3 *D sc + e(3) So what would happen if we just naively run an OLS regression without μ n ? Y = a 0 + a 1 *Z 1 + a 2 *Z 2 + a 3 *D sc + ε (4) What is wrong if we do OLS? ε = μ n + e Cov (Dsc, ε) ≠ 0, (may be + or -)
Omitted variable problem Y = a 0 + a 1 *Z 1 + a 2 *Z 2 + a 3 *D sc + μ n + e Biased and Inconsistent (1) omitted soil quality (a 3 <0 might be due to the fact that worse quality plots are sharecropped), in which case the productivity on owned plot may be higher, this can not be attributed to Marshallian inefficiency (2) omitted farmer’s characteristics (a 3 <0 might be due to the fact that least able farmers or with worse access to working capital sort into sharecropping), again, this can not be attributed to Marshallian inefficiency (3) omitted income opportunities (a 3 >0), a poor sharecropper may have few alternative income opportunities and thus farm the labor more intensively despite the disincentive effect identified by Marshall. Possible Sources of Endogeneity (5)
Remedies Instrumental variable method: find instruments for D sc Fixed effects model (Shaban, 1987 “Testing between Competing Models of Sharecropping” Journal of Political Economy)
So how did Shaban address this? Unlike the naïve model, Shaban was very clever in using a dataset which includes significant number of sharecroppers that also happen to cultivate own land. Therefore, his analysis is able to compare the average input intensities and output per area on owned and sharecropped land of the same household by holding constant household characteristics, such as management, access to non-traded inputs and prices of inputs and outputs
What to test? To test whether landlord is able to effectively monitor tenant’s effort and activities The levels of variable inputs applied to own and sharecropped plots follow: (i) No supervision : (ii) With supervision : (iii) With assumption of F(X,t) to be linearly homogeneous in all inputs and variable inputs (X) are normal, then (i) implies: i=1,…,n. (1) i=1,…,n. (2) for all I=1, …,n.(3) i.e., labor can be monitored … like Cheung … like the case where the landlord “makes” the tenant put on 5 weeks of labor SC is inefficient (marshall is right) 3 weeks … 5 weeks
How to interpret the regression? Step 1: get the set up of the “problem” clearly in your mind Step 2: What is the test: Is Sharecropping Inefficient? Step 3: How large is the inefficiency (last 6 rows, p. 907) … use a “decomposition analysis” defined in equation 9, p. 903
N (whole sample) = 2268 households and 9389 plots n (Shaban’s “tricky” sample) = 352 households and 1420 plots Key to Shaban: He picked a sample of farmers that just happened to be cultivating BOTH their own plot PLUS a Sharecropped plot Owner cultivated plot Sharecropping plot
Basic Set-up of Shaban A system of 8 input equations, i.e., family male labor, family female labor, …, fertilizer and other inputs. Control plot specific effects Control for village effects
Are all of the factors accounted for in the regression table? YES! There are 8 dummy variables (one for each village in the sample (village A to H) Irrigated Area (row 9, p. 906) Plot Value (row 10) 3 Soil Variables (rows 1-3, p. 907) Vil. A Vil. D Vil. E Vil. F Vil. G Vil. B Vil. C Vil. H Why does Shaban include village dummy variables? Because the rules in one village may differ from another … why would that matter? For example: if landowners in Village A provided ½ of fertilizer; but those in village H did not.
Estimation Method For each input equation, the input intensity variable on the left hand side is the difference in the weighted average of that variable on own and sharecropped plots, similarly each of the right hand variables is the difference in the weighted average of that variable on own and sharecropped plots (e.g., Equation 7).
Estimating Impacts of SC on Input Intensity Y = a 0 + a 1 *Z 1 + a 2 *Z 2 + a3*D sc + μ n + e Y = labor/ha Z1 = household effects Z2 = plot effects μ n = unobserved effects Problem? Endogeneity
Estimating Impacts of SC on Input Intensity Y = a 0 + a 1 *Z 1 + a 2 *Z 2 + a3*D sc + μ n + e hh effectsplot effects SC institutions Own plot SC plot
Get rid of OBSERVED & UNOBSERVED EFFECTS by SUBTRACTING (4) – (5)
Equation to be estimated 3 plot-specific variables Constant term Or in the case of India – 7 dummy variables (1/village) Where are the hh effects? Observed AND unobserved?
How to interpret the table? Step 1: get the set up of the “problem clearly in your mind Step 2: What is the test: Is Sharecropping Inefficient? Step 3: How large is the inefficiency (last 6 rows, p. 907) … use a “decomposition analysis” defined in equation 9, p. 903
Test: Is Sharecropping Inefficient? It is on page 906 There is a “separate test for each village” In how many villages is sharecropping inefficient? [in 6 of the 8 villages] [why not in the other 2 … we don’t know, but it could be that all rental is between relatives … that there is very little rental … that villages are small … this is what you want to do in your “field work” … find out in qualitative terms what is happening so you can tell an even “richer” story] Test: joint F-test of all of the coefficients across each row … are they jointly different from zero? [see footnote at bottom of table 3 and discussion of the test on p. XXX in the text]
How to interpret the table? Step 1: get the set up of the “problem clearly in your mind Step 2: What is the test: Is Sharecropping Inefficient? Step 3: How large is the inefficiency (last 6 rows, p. 907) … or what is the magnitude? Use a “decomposition analysis” defined in equation 9, p. 903
Equation (9) from the paper Where 1i, 2i, 3i and 4i are defined as the proportion of the mean difference E(Δx i ) that can be attributed to irrigation, plot value, soil and tenancy, respectively.
ΔX 1 : 29.9 (mean difference) X 1 = family male labor E(X 1 s )=60.1 mandays/ha Owner cultivated plotSharecropping plot E(X 1 o ) = 90 mandays/ha There are four potential sources of differences: Irrigation / soil / value of the land + sharecropping inefficiency …
So for what inputs are the inefficiencies greatest? The reduction of labor on the Sharecropping plot is much greater than fertilizer (which actually is higher on the SC plot versus the OC plot … liquidity constraint?) Why? Because it is difficult (or impossible) to monitor on-farm labor!
Interpreting the decomposition numbers … Total difference (ΔX 1 ) is: 29.9 mandays/ha That due to Sharecropping (SC): 62.5% x 29.9 = 18.7 mandays/ha percent 22.7 8.6 6.2 62.5
ΔX 1 : 18.7 (mean difference) What would be the test, if we did a “real experiment” like an agronomist? E(X 1 s )=71.3 mandays/ha Owner cultivated “part of the plot” Sharecropping “part of the plot” E(X 1 o ) = 90 mandays/ha Then ALL of the difference is due to sharecropping inefficiency … because all plot characteristics are the same … only need to do a “t-test” of the difference in means Or: in regression form: ΔX 1 = α 0 + e (or regress delta X on a constant AND nothing else) X 1 = family male labor
Key Findings of Shaban The difference in input intensities and output on owned minus sharecropped land of the same households (holding plot characteristics constant) are found to be sizeable and significant, suggesting that Sharecropping is indeed “Marshallian Inefficient” The differences also persistent even if sample is restricted to sharecropper-owners who grow a single crop across the two types of plots
Finally, why is it so prevalent if sharecropping is so inefficient? What don’t replace with the alternatives (e.g., fixed wage contract or fixed rent contract)? Why don’t replace with fixed wage contract? –Spatial dispersion; Asymmetric information; and Costly information... so laborers will shirk without extreme amounts of supervision [could and do have piece rates where possible, but for many activities this is not possible... could also make small owner cultivated farms an efficient form of enterprise]... use permanent laborers with above market wage premium... or back load wage payments (or use deposit)
Why is it so prevalent? (cont’d) Why don’t replace with fixed rent contract? –same factors make it impossible to monitor tenant's farming practices, and there is an incentive for tenant to overuse the land (or tractor or bullock [or rental car])... contract will try to be written to protect the fixed asset.. but enforcing contracts are expensive... [often use long term contracts to protect fragile assets]... tenants are usually poor and can not bear risk [or finance the transaction...]
Role of Risk Definition: the uncertainty of the income stream from an activity Who bears it? –Fixed Wage Contract:the landlord bears all... –Fixed Rent Contract:the tenant bears all... –Sharecropping:landlords and tenants share risk...
Explanation for sharecropping? It is a risk sharing institution [but why not give 1/2 wage contract and 1/2 fixed rent contract]... that also seeks to overcome the incentive problem of fixed wage contracts... and usually is done on long term basis or with local landlord (or his/her manager--"the running dog of the landlord") to overcome the incentive problems of misusing the land....
Summary of Progress So Far What is Sharecropping? What is the (theoretical) “problem” with sharecropping? [why did Marshall hate it?] Is it really inefficient? [empirical approach to testing for the inefficiency of SC … answer is “yes”] So why do people choose it? Now the final question is:
Other explanations for sharecropping... joint contract to share specialized skills (or gains from comparative advantage): landlord provides external management skills (e.g., provide information on policy; marketing; new technology); and tenant brings superior internal management skills (e.g., ability to supervise labor-- tenant often employs own family members and uses his/her own tools)... [this might also explain other contractual arrangements outside of agriculture: e.g., the form of management in China' rural enterprises; the form of Jeepney contracts in the Philippines; the practice of subcontracting clothing factories in Mexico, Brazil, and Thailand]
Situations under which sharecropping achieves efficiency (I) The tenant’s work effort can be costlessly enforced by landlords (Johnson, Cheung). (II) Infinitely repeated contracts. Threat or eviction may act as an deterrent to cheating, and cooperation becomes sustainable when the benefits are sufficient and appropriately shared (iii) Individual non-cooperative behavior is identical to the cooperative choice (Rao; Ostuka and Hayami; Arrow, Simon; Sadoulet, Fukui, and Janvry; Heady; Nabi) (iv) Interlinked contracts. Credit transactions, insurance, and sometimes, marketing of the tenant’s product by the landlord are commonly observed complementary contracts between landlord and tenant (OCH)
Final word What is SC? Hopefully you know more about SC than you ever wanted to know Why Theoretically Inefficient? Incentive problem Is it so in reality? Yes (according to Shaban) So why do we observe it? A second best solution Although there is a cost of less than perfect incentives, there are still some incentives; and it helps mitigate risk Helps explain: why we see it in India but not China; why we see it more in the US in the 1980s and less in the 1970s and 1990s … Studying institutions in Development Economics is powerful!
Effective in getting rid of unobserved heterogeneity … Other places that can use: –Panel data –Firm – worker matched data –Etcetera Look at example of irrigation’s impact on yields
Problems with fixed effects In panel data, only controls for part of unobserved effects … those that are NON-TIME VARYING! Can’t answer all questions: what if you are interested in: impact of farmer’s education on yields … can’t use fixed effects … Can be a problem with measurement error … e.g., if there is little variability over time Angus Deaton. The Analysis of Household Surveys: A Mic roeconometric Approach to Development Policy. Published for the World Bank. The Johns Hopkins University Press. Baltimore and London.
Nutrition (ijk) = a0 + a1*conflict (kt-n) + a2*Z (ijkt) + + a3*province + e (ijkt) Cov(conflict, e (ijkt) not