Data analyses

First, we calculated the mean number of specific gears that individual fishers had used throughout their careers. Second, we evaluated how the mean number of specific gears that individual fishers used in a year changed over time. Third, we assessed how two metrics of gear diversity had changed over time for specific gears: gear richness and the Simpson's Index of Diversity [64]. Gear richness (G; hereafter diversity) was estimated as the total number of unique gears (g) used in a year (t)

The Simpson's Index of Diversity (D) considers both gear richness and evenness by estimating the probability that two gears taken at random will represent the same gear. We estimated D where N_t is the total number of any type of gear used by all fishers in year (t) and n_t is the total number of gears of a particular type of gear used by all fishers in a year (t)

For analyzing these gear changes over time, we divided the fishing timelines into the four governance eras and sampled gear-use during six randomly selected years from each era. The residuals from the Bartlett test comparing changes in in gear use and diversity were non-normally distributed, so we used Kruskal-Wallis Rank Sum tests to compare differences in gear use between governance eras [65]. We used a post-hoc Kruskal-Wallis Multiple Comparison test to determine which governance eras exhibited significant differences in gear use and diversity.

To understand trends in fishing activities, we evaluated how three aspects of fishing developed over time and in relation to governance eras: (i) total fishing effort (cumulative number of days fished in one year by male fishers from the 23 participating villages); (ii) relative fishing effort (i.e. the proportion of total fishing effort allocated to each fishing gear in one year); and (iii) the proportion of fishers using various fishing gears in one year. We assessed the evolution of these three aspects of fishing for both the eight general fishing gear categories and the four (non-exclusive) pairs of intensive and non-intensive gears. Estimates for total fishing effort (days per year) assess the fishing effort for only the 23 participating communities.

For each category, we estimated effort and gear use using six steps. First, we used interview data to calculate the total annual fishing effort that all respondents fished as the sum of individual effort (days fished; e) by all respondents (f) in year (t)

Second, we evaluated the mean fishing effort in a year (E-t) as the total days fished (E^t) in a year (t) divided by the total number of respondents (f) in year (t)

Third, for each year we divided individual respondents’ fishing effort among their actively used fishing gears. Fourth, we estimated gear-specific effort as the sum of effort (e) by individual respondents (f) using each gear (g) in each year (t)

Fifth, we estimated the relative (percent) fishing effort allocated to each gear (g) during each year (t) as

Sixth, we estimated the total fishing effort allocated to each gear by multiplying (i) proportion of effort allocated to that fishing gear by all fishers (RE_gt), (ii) mean fishing effort (E-t), (iii) the population of participating villages (V_t), and (iv) the proportion of the population (P_t) who fished during a year (t) (adapted from [8] to include effort and time)

Lacking other data, we assumed that the proportion of the population that fished (P_t) was static through time[48]. Using proportions and estimates rather than raw sums of effort allowed us to compare across the study period, despite the temporally-varying sample sizes of fishers[48].

After obtaining estimates of total fishing effort, relative fishing effort, and the proportion of fishers using various fishing gears, we analyzed changes over time using generalized least square models[65]. The explanatory variables were governance era and year. We assumed governance era accounted for changes in fishing regulations and governance structures. Year may indicate changes in pressures on the ecosystem, underlying changes in the abundance of species, or supporting ecological processes. Models incorporated temporal auto-correlation using a corARMA auto-correlation structure and used governance era as a variance covariate to allow for the heterogeneity of variances within governance eras [65]. Due to high variance in the first decade, we restricted analyses of fishing activity to the period 1960–2010. We conducted all analyses in R 3.3.1 (R Core Team 2016) using packages dplyr, pgirmess, MuMin, and nlme.