Factors Determining Fish Catch in Indonesian Fishing Villages

1
Factors Determining Fish Catch in Indonesian
Fishing Villages

• The data is a subset from a survey of saltwater
  fishermen in the Minahasa region of Northern
  Sulawesi, Indonesia (source of data Professor
  Randall Kramer, Duke University)
• Data includes the following variables
• Average Catch (in Kg/trip)
• Type of Gear Used (nets, hook line, etc)
• Boat ownership
• Type of boat used (canoe, sail, etc)
• Daily fishing habits
• Education level
• Family Size

2
Research Questions

• Which of the factors studied has an effect on the
  fishermens catch size?
• 2) Can we recommend a policy based on this data
  which will reduce overall catch?

3
Data Collection Method

• Systematic sampling from random lists of
  fishermen was used in order to select individuals
  for the survey. A two-stage cluster sampling
  method was used first by sub-districts and second
  by villages within selected sub-districts.
• Raw data was obtained from government statistics
  on population and occupation from 1997 and from
  village chief records (March 1999-June 1999).
• 6 coastal sub-districts (3 from the East and 3
  from the West) were randomly selected out of a
  total of 17 (6 of these were excluded due to
  having less than 5 fishermen). 30 (18) of the
  coastal villages within the selected
  sub-districts were randomly selected for a survey
  using a random number generator.

4
Data Collection Method

• A target of 600 completed interviews was set in
  order to allow sufficient degrees of freedom for
  various econometric analyses. The numbers of
  interviews needed for each village was determined
  (based on the fishermen population size), and the
  interviewers acquired a list of all fishermen
  living in the village. They randomly selected
  fishermen to interview until they had exhausted
  the quota for the village.
• Fishermen in those sub-districts of the Minahasa
  region which had over 5 of the working male
  population primarily engaged in saltwater fishing
  had the same probability of being interviewed.

5

• Analysis of Data Catch vs. Family Size

Hypothesis Testing Let ?1 be the median fish
catch for small family sizeLet ?2 be the median
fish catch for medium family sizeLet ?3 be the
median fish catch for large family size Ho ?1
?2 ?3 HA not all ?i are equal Statistical
tests were run using log transformed data
Categories The original data contained
discrete ordered variables (1 through 9 members
per household) and was pooled into three
categorical groups because of low sample sizes
prior to pooling. Test Assumptions (One-Way
ANOVA)The three groups had relatively equal
variances after log transformations and were
close to normally distributed.
Results There is insufficient evidence to
suggest a difference in the median fish catch
between the small, medium and large family groups
(two-sided p-value of 0.788, one-way analysis of
variance F-test).
6
Analysis of Data Catch vs. Boat Ownership
Hypothesis Testing Let ?0 be the median fish
catch for non boat-ownersLet ?1 be the median
fish catch for boat-owners Ho ?0 ?1 HA ?0 ?
?1 Statistical tests were run using log
transformed data Categories The original data
contained the discrete categorical variables
shown and was not changed. Test Assumptions
(Welchs 2 Sample t-test)Due to unequal
variances after log transformations, a Welchs
Modified t-test was used. The samples were
close to being normally distributed after
transformation.
4 outliers from same village
Summary Statistics (log data)
Results There is insufficient evidence to
suggest a difference in the median fish catch
between the boat owners group and non boat-owners
(two-sided p-value of 0.115, Welch two-sample
t-test).
7
Analysis of Data Catch vs. Boat Type
Hypothesis Testing Let ?1 be the median fish
catch for canoe usersLet ?2 be the median fish
catch for sail boat usersLet ?3 be the median
fish catch for motor boat users Ho ?1 ?2 ?3
HA not all ?i are equal Statistical tests
were run using log transformed data Categories
The original data contained discrete categorical
variables and was not changed. Test Assumptions
(One-Way ANOVA)Given that the sample size in
the group with motor boats is much smaller than
the other two groups, we assumed that its
variance would have been larger had the sample
size been larger. Normality assumptions were met.
Results There is insufficient evidence to
suggest a difference in the median fish catch
between the groups with the three different boat
types (two-sided p-value of 0.053, one-way
analysis of variance F-test).
8
Analysis of Data Catch vs. Education Level
Hypothesis Testing Let ?1 be the median fish
catch for group 1Let ?2 be the median fish catch
for group 2Let ?3 be the median fish catch for
group 3 Ho ?1 ?2 ?3 HA not all ?i are
equal Statistical tests were run using log
transformed data Categories The original data
contained four discrete categorical variables.
The primary and no education variables were
pooled due to a low sample size in the latter.
Test Assumptions (One-Way ANOVA)Given that the
sample sizes in the secondary and high school
education groups were much smaller than the other
two groups, we assumed that their variances would
have been larger had the sample size been larger.
Normality assumptions were met.
Results There is insufficient evidence to
suggest a difference in the median fish catch
between the groups with primary or no education,
secondary and high school education (two-sided
p-value of 0.315, one-way analysis of variance
F-test)
9
Analysis of Data Catch vs. Daily Fishing Habits
Hypothesis Testing Let ?0 be the median fish
catch for non-daily fishermenLet ?1 be the
median fish catch for daily fishermen Ho ?0 ?1
HA ?0 ? ?1 Statistical tests were run using
log transformed data Categories The original
data contained the discrete categorical variables
shown and was not changed. Test Assumptions (2
Sample t-test)The three groups had relatively
equal variances and were close to being normally
distributed after log transformations.
3 outliers from same village (Borgo)
4 outliers from same village (Borgo)
Results There is insufficient evidence to
suggest a difference in the median fish catch
between the group that fishes on a daily basis
and the one that does not (two-sided p-value of
0.59, two-sample t-test).
10
Analysis of Data Catch vs. Gear Type
Hypothesis Testing (K-W Rank Sum Test) Let ?0
be the median fish catch for non daily
fishersLet ?1 be the median fish catch for
daily fishers Ho ?1 ?2 ?3 HA not all ?i
are equal Statistical tests were run using log
transformed data Categories The original data
contained six different types of fishing gear.
The four net types were combined into a single
group due to low sample size in some of the net
subgroups (see next slide). Test AssumptionsDue
to unequal variance in the net group, the
Kruskal-Wallis Rank Sum Test was used. The
Tukey-Kramer pairwise comparison test was run to
determine the difference.
One outlier from Village Sapa
Outliers from village Borgo
Results There is sufficient evidence to suggest
a difference in the median fish catch between at
least two of the three different gear type groups
(two-sided p-value of 0.0002, Kruskal-Wallis Rank
Sum Test). The group with the net gear type had
a median fish catch which was 1.23 to 7.21
kg/trip higher than the spear-fishing group (95
confidence interval,Tukey-Kramer multiple
comparison).
11
Analysis of Data Catch vs. Gear Type
Outlier from village Borgo
Net group consisted of 4 different sub-groups
shown above. Small sample sizes suggested the
need for pooling.
12
Further Analysis of Gear Data
Left Count Table for Gear Type and Average
Catch. Nets seem to account for the majority of
catches in the higher average catch categories,
while Hook and Line fishing accounts for the
majority of counts in the low (0-30) average
catch group.
Above Data in Net group of fishing gear was
comprised of 4 different types of fishing nets.
The data was combined into a single group due to
small sample size in some groups, as seen in the
table to the left. (Note the relative high
abundance of counts for Fly Fishing Net catch
in the 30-50 kg/trip the 50-80 kg/trip
categories.)
13
Further Analysis of Gear Data

• Using count tables to compare all groups to gear
  type, the following was determined
• The majority of fishermen using nets also
• Have a medium-sized family (3-5)
• Low level of education (Primary only)
• Use a canoe
• Fish daily
• Own their own boat
• NOTE Above statements are casual observations
  (not statistical conclusions) from count tables.
  No inferences about the general population can be
  made as a statistical test was not possible for
  these comparisons because of low observations
  (lt5) in some groups.

14
Unique Features of Data

• Types of variables The data consisted of one
  continuous variable (the response variable
  average fish catch) and six categorical
  variables. Some categorical variables were
  created while others were original.
• Outliers All outliers were identified
  throughout the study and had the following
  characteristics all were from the village Borgo
  and did not own boats while one of them (average
  catch 340 Kg/trip) was from the village Sapa
  and owned a boat. It has been suggested that
  fishermen from Borgo are a part of larger
  commercial operations, which may partake in
  multi-day fishing trips.

15

• Shortcomings of the Experimental Design

• Is This Sample Representative of the Larger
  Population? There are doubts as to whether the
  survey sample is representative of the population
  at large as only 6 out of 17 coastal
  sub-districts were selected and from these only
  30 of the coastal villages were sampled. Small
  sample sizes also accounted for difficulty in
  making inferences about the larger population of
  fishermen.
• Accuracy of Preliminary Data? The method used by
  interviewers to acquire the lists of fishermen
  currently living in coastal villages could be
  subject to inaccuracies. Government statistics
  proved to be outdated when compared with actual
  observations. The local team members, therefore,
  resorted to gathering this information by asking
  village chiefs or their staff the names of the
  fishermen currently living in each village.
  There is no guarantee that these figures were
  accurate or that the chiefs were honest when
  sharing this information.
• Accuracy of Survey Methods? Fishermen who
  completed surveys were not guaranteed anonymity
  and this may have influenced the answers they
  provided. Interviewers contacted the fishermen
  they selected from their lists (through
  systematic random sampling) and then asked them
  to complete the survey on the spot. There is no
  indication as to how many fishermen refused to
  fill out the survey or how many were unavailable
  to do so, although the target of 600 completed
  surveys was reached.

16
Conclusions

• Statistical conclusions
• The analysis failed to provide evidence of a
  significant effect on fish catch by all of the
  variables studied with the exception of gear
  type.
• The only statistical difference found was that
  fishermen using nets showed a significantly
  higher catch than spear-fishermen.
• The effect of the other variables on gear type
  could not be determined due to small sample sizes
  and the limited scope of statistical tools
  available for this analysis.
• Implications
• There is not enough evidence based on this study
  to make any policy recommendations, although
  obtaining larger sample sizes may provide
  valuable insight into fishery management
  strategies. Future studies should focus on
  determining the effects of the different gear
  types on average fish catch.