Vietnam

Field work in Vietnam consists of a qualitative assessment with focus group discussions and quantitative surveys. We first look at the consolidated survey results.

Notes:

  • 1 Int’l $ = 7,473.67 VND (Vietnamese Dong) using 2020 World Bank PPP conversion rates (1 Int’l $ = 1 USD)
  • Focus crop = rice
  • Transportation costs are lumped into the cost of pesticides, fertilizers and harvesting.
  • Labor costs are per hectare
  • Inspection and certification fees are per farm (total fees for a single season). Only farmers who sell to seed centers or seed companies do incur these marketing costs.
  • We differentiate between expected yield per ha yield_ha_kg and realized sales in the last season sales_ha_kg.

Survey Recodes

xrate <- 7473.67

# Load respondent data
hh <- fread("../data/vnm/hh.csv")
# Load group data
group <- fread("../data/vnm/group.csv")

There are 21 variables and 60 observations in this set. A summary is shown below.

print(dfSummary(hh), max.tbl.height=500)
Variable Stats / Values Freqs (% of Valid) Valid Missing
Code
[character]		 1. Farmer 1
2. Farmer 10
3. Farmer 11
4. Farmer 12
5. Farmer 13
6. Farmer 14
7. Farmer 15
8. Farmer 16
9. Farmer 17
10. Farmer 18
[ 50 others ]		 1 ( 1.7%)
1 ( 1.7%)
1 ( 1.7%)
1 ( 1.7%)
1 ( 1.7%)
1 ( 1.7%)
1 ( 1.7%)
1 ( 1.7%)
1 ( 1.7%)
1 ( 1.7%)
50 (83.3%)		 60
(100.0%)		 0
(0.0%)
Age
[integer]		 Mean (sd) : 51.7 (10.6)
min < med < max:
32 < 52.5 < 75
IQR (CV) : 15 (0.2)		 32 distinct values 60
(100.0%)		 0
(0.0%)
Sex
[character]		 1. Nam
2. Nữ		 48 (80.0%)
12 (20.0%)		 60
(100.0%)		 0
(0.0%)
Group
[character]		 1. Binh My
2. Ta Ben
3. Trung Hiep
4. Vinh Qui
5. Vinh Trach		 7 (11.7%)
13 (21.7%)
12 (20.0%)
10 (16.7%)
18 (30.0%)		 60
(100.0%)		 0
(0.0%)
Province
[character]		 1. An Giang
2. Bạc Liêu
3. Vĩnh Long		 35 (58.3%)
13 (21.7%)
12 (20.0%)		 60
(100.0%)		 0
(0.0%)
How long have you been a member of this group?
[integer]		 Mean (sd) : 10.2 (7.5)
min < med < max:
1 < 9.5 < 41
IQR (CV) : 11 (0.7)		 18 distinct values 60
(100.0%)		 0
(0.0%)
Cost seed per ha (LCU)
[integer]		 Mean (sd) : 2023800 (636007.9)
min < med < max:
550000 < 1952500 < 3900000
IQR (CV) : 901500 (0.3)		 45 distinct values 60
(100.0%)		 0
(0.0%)
Cost of fertilizer per ha (LCU)
[integer]		 Mean (sd) : 4439145 (1335996)
min < med < max:
2489300 < 4156000 < 8020000
IQR (CV) : 1837800 (0.3)		 59 distinct values 60
(100.0%)		 0
(0.0%)
Cost of pesticide per ha (LCU)
[integer]		 Mean (sd) : 5710617 (2860988)
min < med < max:
810000 < 5157000 < 13212000
IQR (CV) : 4045000 (0.5)		 60 distinct values 60
(100.0%)		 0
(0.0%)
Cost of transport per ha (LCU)
[logical]		 All NA’s 0
(0.0%)		 60
(100.0%)
Labor cost (LCU)
[integer]		 Mean (sd) : 8877388 (2700077)
min < med < max:
3762000 < 8500000 < 15338000
IQR (CV) : 3967000 (0.3)		 58 distinct values 60
(100.0%)		 0
(0.0%)
Inspection / certification Fees (LCU)
[integer]		 Mean (sd) : 67700 (304582.1)
min < med < max:
0 < 0 < 1650000
IQR (CV) : 0 (4.5)		 0 : 56 (93.3%)
300000 : 1 ( 1.7%)
462000 : 1 ( 1.7%)
1650000 : 2 ( 3.3%)		 60
(100.0%)		 0
(0.0%)
Labelling costs per kg (LCU)
[integer]		 1 distinct value 0 : 60 (100.0%) 60
(100.0%)		 0
(0.0%)
Packaging costs per kg (LCU)
[integer]		 Mean (sd) : 2.3 (13.8)
min < med < max:
0 < 0 < 100
IQR (CV) : 0 (5.9)		 0 : 58 (96.7%)
40 : 1 ( 1.7%)
100 : 1 ( 1.7%)		 60
(100.0%)		 0
(0.0%)
Other marketing costs? (LCU)
[integer]		 1 distinct value 0 : 60 (100.0%) 60
(100.0%)		 0
(0.0%)
Estimated Yield (kg/ha)
[integer]		 Mean (sd) : 8763.2 (1628.2)
min < med < max:
6300 < 8480 < 13000
IQR (CV) : 2500 (0.2)		 25 distinct values 60
(100.0%)		 0
(0.0%)
Selling price of seed per kg (LCU)
[integer]		 Mean (sd) : 7572.5 (2001.3)
min < med < max:
5300 < 7000 < 13000
IQR (CV) : 600 (0.3)		 19 distinct values 60
(100.0%)		 0
(0.0%)
How many kg were sold in the season?
[integer]		 Mean (sd) : 8506.8 (1776.3)
min < med < max:
3200 < 8200 < 12000
IQR (CV) : 2850 (0.2)		 29 distinct values 60
(100.0%)		 0
(0.0%)
What was your expected gross margin?
[numeric]		 Mean (sd) : 0.6 (0.1)
min < med < max:
0.2 < 0.7 < 0.8
IQR (CV) : 0.1 (0.2)		 33 distinct values 60
(100.0%)		 0
(0.0%)
Total production cost
[integer]		 Mean (sd) : 21138367 (4025039)
min < med < max:
12465300 < 21671900 < 33338000
IQR (CV) : 4829735 (0.2)		 60 distinct values 60
(100.0%)		 0
(0.0%)
Gross sales
[integer]		 Mean (sd) : 63491772 (16928812)
min < med < max:
28329000 < 63660000 < 97500000
IQR (CV) : 24325000 (0.3)		 49 distinct values 60
(100.0%)		 0
(0.0%)

Recode variable names (see codebook).

setnames(hh, lbl$label, lbl$code, skip_absent=T)

Additional recodes for categorical variables. Note that we create a categorical variable ssd to indicate whether a farmer currently engages in formal seed system distribution. For consistency across countries we also reclassify age into 2 categories < 30 and ≥ 30.

setorder(hh, adm1_nm, group, gender)

hh[, `:=`(
  hhid = paste("VNM", gsub(" ", "0", format(1:.N, width=3)), sep=""),
  iso3 = "VNM",
  crop = "rice",
  adm1_nm = factor(adm1_nm),
  group = factor(group, levels=hh[, unique(group)]),
  gender = factor(gender, levels=c("Nam", "Nữ"), labels=c("Male", "Female")),
  ssd = factor(cert_lcu > 0, levels=c(F, T), labels=c("Informal", "Formal")),
  age_num = age,
  age = factor(age >= 30, levels=c(F, T), labels=c("< 30", "≥ 30")),
  years = factor(member_years >= 5, levels=c(F, T), labels=c("< 5", "≥ 5"))
)]

Spatial Covariates

Using community GPS coordinates we also suggest to enrich this dataset with additional biophysical and geospatial variables, e.g.:

  • Agroecological zone
  • Travel time to nearest market
  • Distance to nearest seed center / company
  • Size of nearest seed center / company
  • Population density
  • Last season total rainfall
  • Last season heat stress days (if any)

[pending GPS coordinates]

Constructed Variables

Farmers report both expected yields yield_ha_kg and actual sales in the last season sales_ha_kg, so we can construct both expected and realized costs in monetary terms costs_exp_ha_lcu and costs_real_ha_lcu. Note that we then use realized sales to calculate profitability metrics.

hh[, 
  tran_ha_lcu := as.numeric(tran_ha_lcu)
][, 
  tran_ha_lcu := fifelse(is.na(tran_ha_lcu), 0, tran_ha_lcu)
][, `:=`(
  # Expected costs
  costs_exp_ha_lcu = 
    # Per ha costs
    seed_ha_lcu + fert_ha_lcu + pest_ha_lcu + tran_ha_lcu + labor_ha_lcu + cert_lcu +
    # Per kg costs
    yield_ha_kg * (labl_kg_lcu + pckg_kg_lcu + mark_kg_lcu),  
  
  # Realized costs
  costs_real_ha_lcu = 
    # Per ha costs
    seed_ha_lcu + fert_ha_lcu + pest_ha_lcu + tran_ha_lcu + labor_ha_lcu + cert_lcu +
    # Per kg costs
    sales_ha_kg * (labl_kg_lcu + pckg_kg_lcu + mark_kg_lcu)
)]

Using realized costs and sales, we construct gross margin per ha margin_ha_lcu, total sales sales_ha_sh and profit margin margin_ha_sh per unit of (variable) input costs, and costs_ha_ppp, sales_ha_ppp and margin_ha_ppp in PPP terms to allow for comparisons across groups and countries.

We also construct a measure of total factor productivity tfp as expected output per unit of (expected) input costs. Strictly speaking it is only “partial factor productivity” here because we don’t include the rental cost of land, land preparation costs, irrigation costs, and the costs of animal and mechanical implements.

hh[, `:=`(
  sales_exp_ha_lcu = yield_ha_kg * sales_kg_lcu,
  sales_real_ha_lcu = sales_ha_kg * sales_kg_lcu
)][, `:=`(
  margin_ha_lcu = sales_real_ha_lcu - costs_real_ha_lcu
)][, `:=`(
  # Shares
  sales_ha_sh = sales_real_ha_lcu / costs_real_ha_lcu,
  margin_ha_sh = margin_ha_lcu / costs_real_ha_lcu,
  # PPP$
  costs_ha_ppp = costs_real_ha_lcu / xrate,
  sales_ha_ppp = sales_real_ha_lcu / xrate,
  margin_ha_ppp = margin_ha_lcu / xrate
)][, `:=`(
  tfp = yield_ha_kg / (costs_exp_ha_lcu / xrate)
)]

Below we append some of the information that was recorded at the group level.

kbl(group, align="lccccccc")
Group Established Members Soil Seasons Irrigation Market access Transboundary trade
Ta Ben 2001 30 loamy 2.0 good Vicinity to local market, good road to infrastructure No
Trung Hiep 2003 8 sandy-silty 2.5 good Vicinity to local market, good road to infrastructure No
Vinh Trach 2004 15 clay 3.0 good Vicinity to local market, good road to infrastructure Yes
Binh My 2004 8 clay 3.0 good Vicinity to local market, good road to infrastructure Yes
Vinh Qui 2002 40 clay 3.0 good Vicinity to local market, good road to infrastructure Yes 
# Merge
hh[group, on=.(group=Group), `:=`(
  group_year = `Established`,
  group_size = `Members`,
  soil_type = `Soil`,
  seasons = `Seasons`,
  irrigated = `Irrigation`,
  market_access = `Market access`,
  ttrade = `Transboundary trade`
)]

Finally we normalize all farmer cost line items into a “long” table hh_prod_cost for charting.

# Normalize production cost table
hh_prod_cost <- hh[, .(hhid,
  Seeds = seed_ha_lcu, 
  Fertilizer = fert_ha_lcu, 
  Pesticides = pest_ha_lcu, 
  Labor = labor_ha_lcu,
  Transport = tran_ha_lcu, 
  Certification = cert_lcu,
  Labeling = sales_ha_kg * labl_kg_lcu,
  Packaging = sales_ha_kg * pckg_kg_lcu,
  Marketing = sales_ha_kg * mark_kg_lcu
)]

hh_prod_cost <- melt(hh_prod_cost, id.vars=1, value.name="lcu", variable.name="type")

And we lump all marketing costs into a single category.

levels(hh_prod_cost$type) <- levels(hh_prod_cost$type)[c(1,2,3,4,9,9,9,9,9)]

hh_prod_cost <- hh_prod_cost[, .(
  lcu = sum(lcu, na.rm=T)
), by=.(hhid, type)
][, `:=`(
  # Add cost shares and PPP terms
  share = lcu/sum(lcu, na.rm=T),
  ppp = lcu/xrate
), by=.(hhid)
][hh, on=.(hhid), `:=`(
  # Add categorical variables
  group = group,
  gender = gender,
  age = age,
  years = years,
  crop = crop,
  ssd = ssd
)]

Note that in the current survey we are missing farm sizes (or planted acreage), so we can not directly study the effect of farm size on the per-unit costs of production and yields, or look at potential scale effects on a farmer’s efficiency and profitability. We can however study whether larger groups might have positive effects.

Descriptive Statistics

Respondent Characteristics

Breakdown by categorical variables.

ggplot(
  hh[, .N, by=.(group, age, gender, crop, years)],
  aes(axis1=crop, axis2=gender, axis3=age, axis4=years, y=N)) +
  geom_alluvium(aes(fill=group), width=1/4, alpha=.7, color="white") +
  geom_stratum(width=1/4) +
  geom_text(stat="stratum", aes(label=after_stat(stratum)), angle=90, size=2.2) +
  scale_x_discrete(limits=c("Crop", "Gender", "Age", "Years in Seed Club")) +
  labs(y=NULL, fill="Seed Club",
    title = "Categories of Survey Respondents - Vietnam",
    subtitle = "Stratified by seed club") +
  theme_def(axis.text=element_text(face="bold"))

Showing contingency tables between each pair of categorical variables (seed club group, gender, years in seed club years, and use of formal seed system distribution ssd). Rice in Vietnam is a male-dominated production, hence the absence of female respondents in a few clubs.

ttt_ftable(hh, vars=c("group", "gender", "years"))
Contingency Table (% of respondents)
group gender < 5 ≥ 5 Sum

N = 60 | Mantel-Haenszel chi-squared = 18.41 | p-value = 0.0010
Binh My Male 5 6.7 11.7
Sum 5 6.7 11.7
Vinh Qui Male 5 11.7 16.7
Sum 5 11.7 16.7
Vinh Trach Male 11.7 16.7 28.3
Female 1.7 0 1.7
Sum 13.3 16.7 30
Ta Ben Male 0 16.7 16.7
Female 0 5 5
Sum 0 21.7 21.7
Trung Hiep Male 0 6.7 6.7
Female 0 13.3 13.3
Sum 0 20 20
Sum Male 21.7 58.3 80
Female 1.7 18.3 20
Sum 23.3 76.7 100

Seed Production Costs

General breakdown and distribution of (realized) input costs across seed clubs, gender, and input type.

ttt(costs_ha_ppp ~ group | gender, data=hh, render=fmt,
  caption="Total Input Costs in Absolute Terms (PPP$ / ha) - Vietnam")
Total Input Costs in Absolute Terms (PPP$ / ha) - Vietnam
group Statistic Male Female
Binh My mean 2,913 NA
median 3,106 NA
sd 642 NA
Vinh Qui mean 3,011 NA
median 2,950 NA
sd 621 NA
Vinh Trach mean 2,882 2,520
median 2,938 2,520
sd 442 NA
Ta Ben mean 2,817 2,076
median 2,806 2,048
sd 447 235
Trung Hiep mean 2,401 2,961
median 2,437 2,973
sd 685 481

Boxplots with mean comparison p-value and significance levels. Each level is compared to the sample mean.

(ns : p > 0.05, * : p ≤ 0.05, ** : p ≤ 0.01, *** = p ≤ 0.001, **** = p ≤ 0.0001)

ggBoxTest(hh, aes(gender, costs_ha_ppp, fill=gender, color=gender), cp=list(1:2)) +
  scale_y_continuous(labels=comma) +
  facet_wrap(~crop) +
  labs(x="", y="", fill="",
    title="Total Input Costs (PPP$ / ha) - Vietnam",
    subtitle="Stratified by gender") +
  theme_def(legend.position="none")


ggBoxTest(hh, aes(group, costs_ha_ppp, fill=group, color=group), ref=".all.") +
  facet_wrap(~crop) +
  scale_y_continuous(labels=comma) +
  labs(x="", y="", fill="",
    title="Total Input Costs (PPP$ / ha) - Vietnam",
    subtitle="Stratified by seed club") +
  theme_def(legend.position="none")

Breakdown across categories of farm input.

ttt(ppp ~ type | gender, data=hh_prod_cost, render=fmt,
  caption="Input Costs in Absolute Terms by Gender (PPP$ / ha) - Vietnam")
Input Costs in Absolute Terms by Gender (PPP$ / ha) - Vietnam
type Statistic Male Female
Seeds mean 281 228
median 270 220
sd 91 34
Fertilizer mean 613 517
median 592 496
sd 179 161
Pesticides mean 748 827
median 690 718
sd 385 384
Labor mean 1,203 1,128
median 1,136 1,167
sd 391 205
Marketing mean 14 3
median 0 0
sd 49 12 
tbl <- hh_prod_cost[, .(
  ppp = mean(ppp, na.rm=T)
), keyby=.(gender, ssd, type)]

ggplot(tbl, aes(gender, ppp, fill=type)) +
  geom_bar(stat="identity", position="fill", alpha=.7, width=.6, color="white") +
  scale_y_continuous(labels=percent) +
  facet_wrap(~ssd) +
  labs(y="", x="",
    title="Breakdown of Input Costs by Category - Vietnam",
    subtitle="Stratified by gender and seed system") +
  theme_def(legend.position="right")

ttt(ppp ~ type | years, data=hh_prod_cost, render=fmt,
  caption="Input Costs in Absolute Terms by Years in Seed Group (PPP$ / ha) - Vietnam")
Input Costs in Absolute Terms by Years in Seed Group (PPP$ / ha) - Vietnam
type Statistic < 5 ≥ 5
Seeds mean 284 267
median 287 259
sd 64 91
Fertilizer mean 648 578
median 616 538
sd 183 176
Pesticides mean 810 750
median 721 680
sd 363 391
Labor mean 1,147 1,200
median 1,080 1,167
sd 406 351
Marketing mean 0 15
median 0 0
sd 0 50 
tbl <- hh_prod_cost[, .(
  ppp = mean(ppp, na.rm=T)
), keyby=.(years, crop, type)]

ggplot(tbl, aes(years, ppp, fill=type)) +
  geom_bar(stat="identity", position="fill", alpha=.7, width=.6, color="white") +
  scale_y_continuous(labels=percent) +
  facet_wrap(~crop) +
  labs(y="", x="",
    title="Breakdown of Input Costs by Category - Vietnam",
    subtitle="Stratified by crop and years in seed club") +
  theme_def(legend.position="right")

ttt(ppp ~ type | ssd, data=hh_prod_cost, render=fmt,
  caption="Input Costs in Absolute Terms by Seed System Type (PPP$ / ha) - Vietnam")
Input Costs in Absolute Terms by Seed System Type (PPP$ / ha) - Vietnam
type Statistic Informal Formal
Seeds mean 265 347
median 261 311
sd 80 132
Fertilizer mean 597 551
median 563 507
sd 179 196
Pesticides mean 781 531
median 697 486
sd 384 322
Labor mean 1,167 1,480
median 1,123 1,452
sd 359 277
Marketing mean 2 149
median 0 168
sd 14 88 
tbl <- hh_prod_cost[, .(
  ppp = mean(ppp, na.rm=T)
), keyby=.(crop, group, type)]

ggplot(tbl, aes(group, ppp, fill=type)) +
  geom_bar(stat="identity", position="fill", alpha=.7, width=.6, color="white") +
  scale_y_continuous(labels=percent) +
  facet_wrap(~crop) +  
  labs(y="", x="",
    title="Breakdown of Input Costs by Category - Vietnam",
    subtitle="Stratified by seed club") +
  theme_def(legend.position="right")

Are there significant differences across groups? We first compare input cost shares across gender, then across seed clubs.

ggBoxTest(hh_prod_cost, 
  aes(type, share, fill=gender, color=gender), 
  grp.c=aes(group=type), grp.s=aes(group=gender)) +
  scale_y_continuous(labels=percent) +    
  facet_wrap(~crop) +  
  labs(x="", y="", fill="", color="",
    title="Input Costs by Category (Percent of Total Costs by Ha) - Vietnam",
    subtitle="Stratified by gender") +
  theme_def(legend.position="top")  

ggBoxTest(hh_prod_cost, 
  aes(type, share, fill=group, color=group),
  grp.c=aes(group=type), grp.s=aes(group=group)) +
  scale_y_continuous(labels=percent) +  
  facet_wrap(~crop) +  
  labs(x="", y="", fill="", color="",
    title="Input Costs by Category (PPP$ by Hectare) - Vietnam",
    subtitle="Stratified by seed club") +
  theme_def(legend.position="top")

Efficiency

Differences in productivity measures (expected seed yields and sales) across groups.

ttt(yield_ha_kg ~ group | gender+crop, data=hh, render=fmt,
  caption="Expected Rice Seed Yield (kg / ha) - Vietnam")
Expected Rice Seed Yield (kg / ha) - Vietnam
group Statistic rice
Male Female
Binh My mean 8,272 NA
median 8,400 NA
sd 1,325 NA
Vinh Qui mean 9,850 NA
median 10,250 NA
sd 1,717 NA
Vinh Trach mean 8,709 9,000
median 8,460 9,000
sd 1,883 NA
Ta Ben mean 9,770 7,467
median 10,000 7,700
sd 1,113 404
Trung Hiep mean 7,182 7,938
median 7,250 7,750
sd 623 904 
ttt(sales_ha_kg ~ group | gender+crop, data=hh, render=fmt,
  caption="Seed Sales (kg / ha) - Vietnam")
Seed Sales (kg / ha) - Vietnam
group Statistic rice
Male Female
Binh My mean 8,272 NA
median 8,400 NA
sd 1,325 NA
Vinh Qui mean 9,620 NA
median 10,250 NA
sd 2,040 NA
Vinh Trach mean 8,155 9,000
median 8,000 9,000
sd 1,788 NA
Ta Ben mean 9,770 7,313
median 10,000 7,238
sd 1,113 356
Trung Hiep mean 6,382 7,938
median 7,250 7,750
sd 2,149 904 
ttt(yield_ha_kg ~ group | years+crop, data=hh, render=fmt,
  caption="Realized Seed Sales (kg / ha) - Vietnam")
Realized Seed Sales (kg / ha) - Vietnam
group Statistic rice
< 5 ≥ 5
Binh My mean 6,948 9,265
median 6,923 9,330
sd 45 666
Vinh Qui mean 7,833 10,714
median 8,000 11,000
sd 289 1,220
Vinh Trach mean 8,850 8,626
median 9,100 8,230
sd 1,842 1,911
Ta Ben mean NA 9,238
median NA 9,300
sd NA 1,406
Trung Hiep mean NA 7,686
median NA 7,500
sd NA 874 
ttt(sales_ha_kg ~ group | years+crop, data=hh, render=fmt,
  caption="Realized Seed Sales (kg / ha) - Vietnam")
Realized Seed Sales (kg / ha) - Vietnam
group Statistic rice
< 5 ≥ 5
Binh My mean 6,948 9,265
median 6,923 9,330
sd 45 666
Vinh Qui mean 7,833 10,386
median 8,000 11,000
sd 289 1,984
Vinh Trach mean 8,850 7,683
median 9,100 7,600
sd 1,842 1,565
Ta Ben mean NA 9,203
median NA 9,300
sd NA 1,453
Trung Hiep mean NA 7,419
median NA 7,500
sd NA 1,538

Differences in efficiency measures across gender with mean comparison (Wilcoxon) p-value.

ggBoxTest(hh, aes(gender, yield_ha_kg, color=gender, fill=gender), cp=list(1:2)) +
  scale_y_continuous(labels=comma) +
  labs(x="", y="", fill="",
    title="Expected Rice Seed Yield (kg / ha) - Vietnam",
    subtitle="Stratified by gender") +
  theme_def(legend.position="none")

ggBoxTest(hh, aes(gender, sales_ha_ppp, fill=gender), cp=list(1:2)) +
  scale_y_continuous(labels=comma) +
  labs(x="", y="", fill="",
    title="Total Seed Sales (PPP$ / ha) - Vietnam",
    subtitle="Stratified by gender") +
  theme_def(legend.position="none")

Differences in efficiency measures by years in seed club with mean comparison (Wilcoxon) p-value.

ggBoxTest(hh, aes(years, yield_ha_kg, color=years, fill=years), cp=list(1:2)) +
  scale_y_continuous(labels=comma) +
  labs(x="", y="", fill="",
    title="Expected Seed Yield (kg / ha) - Vietnam",
    subtitle="Stratified by years in seed club") +
  theme_def(legend.position="none")

ggBoxTest(hh, aes(years, sales_ha_ppp, color=years, fill=years), cp=list(1:2)) +
  scale_y_continuous(labels=comma) +
  labs(x="", y="", fill="",
    title="Total Seed Sales (PPP$ / ha) - Vietnam",
    subtitle="Stratified by years in seed club") +
  theme_def(legend.position="none")

Differences in efficiency measures across seed clubs with global ANOVA p-value.

ggBoxTest(hh, aes(group, yield_ha_kg, color=group, fill=group)) +
  scale_x_discrete(labels=label_wrap(5)) +
  scale_y_continuous(labels=comma) +  
  labs(x="", y="", fill="",
    title="Rice Seed Yield (Kg / ha) - Vietnam",
    subtitle="Stratified by seed club") +
  theme_def(legend.position="none")

ggBoxTest(hh, aes(group, sales_ha_ppp, color=group, fill=group)) +
  scale_x_discrete(labels=label_wrap(5)) +
  scale_y_continuous(labels=comma) +  
  labs(x="", y="", fill="",
    title="Total Seed Sales (PPP$ / ha) - Vietnam",
    subtitle="Stratified by seed club") +
  theme_def(legend.position="none")

Looking at production frontiers (units of output vs. units of input). We expect S-shape curves with farmers at different levels of technical efficiency along the curve.

Note that Farmer VNM013 in Winh Qui has total costs over PPP$ 4,000/ha. He was excluded from the approximated curves below.

outlier <- hh[costs_ha_ppp > median(costs_ha_ppp) + 3*sd(costs_ha_ppp), hhid]

kbl(
  caption="Farmers with total input costs > median + 3*sd",
  hh[hhid %in% outlier, .(hhid, group, crop, yield_ha_kg, costs_ha_ppp)],
  format.args=list(big.mark=",", digits=0))
Tab. 2: Farmers with total input costs > median + 3*sd
hhid group crop yield_ha_kg costs_ha_ppp 
ggplot(hh[!hhid %in% outlier], aes(costs_ha_ppp, yield_ha_kg)) +
  geom_smooth(size=.8) +
  geom_point(alpha=.7, shape=20, color=1) +
  scale_x_continuous(labels=comma) +
  scale_y_continuous(labels=comma) +
  labs(x="", y="",
    title="Production Frontier (Output vs. Input) - Vietnam",
    subtitle="Each point is a respondent. Shade shows 90% (kg vs. PPP$ / ha)") +
  theme_def(legend.position="none")

ggplot(hh[!hhid %in% outlier], aes(costs_ha_ppp, yield_ha_kg)) +
  geom_smooth(aes(color=gender, fill=gender), size=.8, level=.9) +
  geom_point(alpha=.7, shape=20) +
  scale_x_continuous(labels=comma) +
  scale_y_continuous(labels=comma) +
  facet_wrap(~gender, scales="free_x") +
  labs(x="", y="",
    title="Production Frontier (Output vs. Input) - Vietnam",
    subtitle="Each point is a respondent. Shade shows 90% CI (kg vs. PPP$ / ha)") +
  theme_def(legend.position="none")

ggplot(hh[!hhid %in% outlier], aes(costs_ha_ppp, yield_ha_kg)) +
  geom_smooth(aes(color=group, fill=group), size=.8, level=.9) +
  geom_point(alpha=.7, shape=20) +
  scale_x_continuous(labels=comma) +
  scale_y_continuous(labels=comma) +
  facet_wrap(~group) +
  coord_cartesian(ylim=c(4000, 14000)) +
  labs(x="", y="",
    title="Production Frontier (Output vs. Input) - Vietnam",
    subtitle="Each point is a respondent. Shade shows 90% CI (kg vs. PPP$ / ha)") +
  theme_def(legend.position="none")

Profitability

Farmers’ gross profit margins by gender and years in seed club.

ttt(margin_ha_ppp ~ group | gender+years, data=hh, render=fmt,
  caption="Mean Gross Profit Margin in Absolute Terms (PPP$ / ha) - Vietnam")
Mean Gross Profit Margin in Absolute Terms (PPP$ / ha) - Vietnam
group Statistic < 5 ≥ 5
Male Female Male Female
Binh My mean 2,971 NA 4,491 NA
median 3,427 NA 4,074 NA
sd 869 NA 1,398 NA
Vinh Qui mean 3,773 NA 6,610 NA
median 4,144 NA 6,768 NA
sd 1,072 NA 1,548 NA
Vinh Trach mean 4,889 6,391 3,946 NA
median 5,405 6,391 4,133 NA
sd 1,968 NA 1,940 NA
Ta Ben mean NA NA 6,627 5,003
median NA NA 6,600 4,828
sd NA NA 925 596
Trung Hiep mean NA NA 7,084 8,235
median NA NA 7,648 8,665
sd NA NA 2,657 1,947 
ttt(margin_ha_sh ~ group | gender+years, data=hh, render=fmt_pct,
  caption="Mean Gross Profit Margin in Relative Terms (% of total input costs) - Vietnam")
Mean Gross Profit Margin in Relative Terms (% of total input costs) - Vietnam
group Statistic < 5 ≥ 5
Male Female Male Female
Binh My mean 127% NA 153% NA
median 116% NA 130% NA
sd 74% NA 73% NA
Vinh Qui mean 136% NA 222% NA
median 163% NA 220% NA
sd 69% NA 41% NA
Vinh Trach mean 162% 254% 148% NA
median 169% 254% 152% NA
sd 49% NA 87% NA
Ta Ben mean NA NA 240% 245%
median NA NA 222% 236%
sd NA NA 48% 56%
Trung Hiep mean NA NA 304% 285%
median NA NA 256% 249%
sd NA NA 136% 91% 
ggplot(hh, aes(x=hhid, color=group)) +
  geom_hline(aes(yintercept=0), color=1) +
  geom_linerange(aes(ymin=0, ymax=margin_ha_ppp), size=.6) +
  geom_point(aes(y=0), shape=20, size=1.4) +
  geom_point(aes(y=margin_ha_ppp, shape=margin_ha_ppp < 0, fill=group), size=1.4) +
  scale_y_continuous(labels=comma) +
  scale_shape_manual(values=24:25) +
  guides(x="none", shape="none") +
  labs(x=NULL, y=NULL, color="", fill="",
    title="Profit Margin (PPP$ / ha) - Vietnam",
    subtitle="Each bar is a respondent's gross profit margin") +
  theme_def(
    legend.position="right",
    panel.grid.major.x=element_blank()
  )

Farmers’ gross profit margins by gender and across seed clubs in both absolute terms and in relative terms as percentage of total input costs per hectare.

ggBoxTest(hh, aes(gender, margin_ha_ppp, color=gender, fill=gender), cp=list(1:2)) +
  scale_y_continuous(labels=comma) +
  labs(x="", y="", fill="",
    title="Gross Profit Margin in Absolute Terms - Vietnam",
    subtitle="Stratified by gender (PPP$ / ha)") +
  theme_def(legend.position="none")

ggBoxTest(hh, aes(gender, margin_ha_sh, color=gender, fill=gender), cp=list(1:2)) +
  scale_y_continuous(labels=percent) +
  labs(x="", y="", fill="",
    title="Gross Profit Margin in Relative Terms - Vietnam",
    subtitle="Stratified by gender (% of total costs)") +
  theme_def(legend.position="none")

ggBoxTest(hh, aes(years, margin_ha_ppp, color=years, fill=years), cp=list(1:2)) +
  scale_y_continuous(labels=comma) +
  labs(x="", y="", fill="",
    title="Gross Profit Margin in Absolute Terms - Vietnam",
    subtitle="Stratified by years in seed club (PPP$ / ha)") +
  theme_def(legend.position="none")

ggBoxTest(hh, aes(years, margin_ha_sh, color=years, fill=years), cp=list(1:2)) +
  scale_y_continuous(labels=percent) +
  labs(x="", y="", fill="",
    title="Gross Profit Margin in Relative Terms - Vietnam",
    subtitle="Stratified by years in seed club (% of total costs)") +
  theme_def(legend.position="none")

ggBoxTest(hh, aes(group, margin_ha_ppp, color=group, fill=group)) +
  scale_x_discrete(labels=label_wrap(5)) +
  scale_y_continuous(labels=comma) +  
  labs(x="", y="", fill="",
    title="Gross Profit Margin in Absolute Terms - Vietnam",
    subtitle="Stratified by seed club (PPP$ / ha)") +
  theme_def(legend.position="none")

ggBoxTest(hh, aes(group, margin_ha_sh, color=group, fill=group)) +
  scale_x_discrete(labels=label_wrap(5)) +  
  scale_y_continuous(labels=percent) +  
  labs(x="", y="", fill="",
    title="Gross Profit Margin in Relative Terms - Vietnam",
    subtitle="Stratified by seed club (% of total costs)") +
  theme_def(legend.position="none")

ggplot(hh[!hhid %in% outlier], aes(member_years, margin_ha_ppp)) +
  geom_smooth(size=.8) +
  geom_point(alpha=.7, shape=20) +
  scale_x_continuous(limits=c(0, 22)) +
  scale_y_continuous(labels=comma) +
  labs(x="", y="", color="",
    title="Gross Profit Margin in Absolute Terms vs. Years in Seed Club - Vietnam",
    subtitle="Each point is a respondent (years vs. PPP$)") +
  theme_def(legend.position="top")

Correlation

Significant pairwise associations.

ggpairs(
  hh[, .(`seed club`=group, `age`=age_num, `years in club`=member_years,
    `costs PPP$`=costs_ha_ppp, `seed yield kg/ha`=yield_ha_kg,
    `margin PPP$`=margin_ha_ppp, `margin %`=margin_ha_sh)],
  upper = list(
    continuous=wrap("cor", size=4), 
    combo=wrap("summarise_by", color=pal[1:5], size=2)),
  lower = list(
    continuous=wrap("smooth", shape=NA), 
    combo=wrap("box_no_facet", fill=pal[1:5], alpha=.8)),
  diag = list(
    continuous=wrap("densityDiag", fill=NA),
    discrete=wrap("barDiag", fill=pal[1:5], alpha=.8)),
  title="Correlogram stratified by seed club - Vietnam"
) + 
  theme_def(
    strip.text=element_text(hjust=.5),
    axis.text.x=element_text(angle=-45),
    panel.grid.major=element_blank()
  )

ggpairs(
  hh[, .(gender, `age`=age_num, `years in club`=member_years,
    `costs PPP$`=costs_ha_ppp, `seed yield kg/ha`=yield_ha_kg, 
    `margin PPP$`=margin_ha_ppp, `margin %`=margin_ha_sh)],
  upper = list(
    continuous=wrap("cor", size=4), 
    combo=wrap("summarise_by", color=pal[1:2], size=2)),
  lower = list(
    continuous=wrap("smooth", shape=NA, color=hh[, pal[gender]]), 
    combo=wrap("box_no_facet", fill=pal[1:2], alpha=.8)),
  diag = list(
    continuous=wrap("densityDiag", fill=NA),
    discrete=wrap("barDiag", fill=pal[1:2], alpha=.8)),
  title="Correlogram stratified by gender - Vietnam"
) +   
  theme_def(
    strip.text=element_text(hjust=.5),
    panel.grid.major=element_blank()
  )

ggpairs(
  hh[, .(`years in club`=years, `age`=age_num,
    `costs PPP$`=costs_ha_ppp, `seed yield kg/ha`=yield_ha_kg, 
    `margin PPP$`=margin_ha_ppp, `margin %`=margin_ha_sh)],
  upper = list(
    continuous=wrap("cor", size=4), 
    combo=wrap("summarise_by", color=pal[1:2], size=2)),
  lower = list(
    continuous=wrap("smooth", shape=NA, color=hh[, pal[gender]]), 
    combo=wrap("box_no_facet", fill=pal[1:2], alpha=.8)),
  diag = list(
    continuous=wrap("densityDiag", fill=NA),
    discrete=wrap("barDiag", fill=pal[1:2], alpha=.8)),
  title="Correlogram stratified by years in seed club - Vietnam"
) +   
  theme_def(
    strip.text=element_text(hjust=.5),
    panel.grid.major=element_blank()
  )

saveRDS(hh, "../tmp/data_vnm.rds")