Wetzel et al.: The effect of reduced data on monitoring overfished fish stocks 
205 
individual are likely more complex with extended peri¬ 
ods of high or low mortality that is affected by external 
factors (e.g., predator abundance, climate conditions)— 
periods that could result in large biases in estimated 
quantities if they are not accounted for in an assess¬ 
ment (Johnson et ah, 2015). 
Shifts in the form of selectivity over time and the 
impact of annual deviations in selectivity led to mixed 
results. The estimation method consistently overesti¬ 
mated the mean size at maximum selectivity for all 
data scenarios with time-varying selectivity. The op¬ 
erating model selectivity applied normally distributed 
deviations to generate the annual shifts in selectivity. 
One would not a priori predict the estimation method 
to have a consistent bias in estimates; however, the es¬ 
timation method was able to identify the change in the 
selectivity form (asymptotic to dome-shaped through a 
reduction in the width at peak selectivity) during the 
rebuilding years with a similar error to that observed 
in the time-invariant case. Each case led to estimates 
that overestimated the width at maximum selectivity, 
the parameter defining the dome in selectivity (dome¬ 
shaped selectivity occurring at larger sizes with in¬ 
creased sizes subject to full selectivity compared with 
that in the operating model). Time blocks were applied 
within the estimation method defined by the status of 
the stock to allow shifts in selectivity, ignoring the an¬ 
nual deviations in the selectivity curve. Studies have 
evaluated other ways of estimating time-varying selec¬ 
tivity by using state-space models (Nielsen and Berg, 
2014) or have examined the implications of applying 
time blocks versus allowing a random-walk component 
in selectivity parameters or catchability (Wilberg and 
Bence, 2006; Martell and Stewart, 2014). Further ex¬ 
ploration should be conducted to determine whether 
allowing a random walk or applying an alternative es¬ 
timation method eliminates the bias detected in the es¬ 
timated selectivity observed here and how data quan¬ 
tity and quality affect these estimates. Additionally, if 
shifts in fishery selectivity are anticipated as a result 
of management actions, increased data collection may 
be required to achieve a similar level of precision in 
estimates of fishery selectivity during rebuilding. 
As with other simulation studies, simplifying as¬ 
sumptions were used in this study and these can lead 
to an underestimation of the uncertainty that would be 
expected in a real-world population. With the estima¬ 
tion method used in this study, the population struc¬ 
ture and functional form of biological relationships 
were assumed correctly—variables that are not known 
with certainty for a typical assessment. Additionally, 
the simulated composition data from the historical and 
management periods were representative of a homog¬ 
enous population. In reality, one may expect spatial 
structure in fish populations, and, during a period of 
limited sampling, composition data may be available 
only from a subset of the population that may not be 
representative of the population as a whole. The re¬ 
sults from this simulation study should be considered a 
best-case scenario specifically designed to allow clearer 
interpretation of the results regarding the availability 
of data for estimate rebuilding. 
The work described here highlights the benefits of 
continued data collection during stock rebuilding on 
the precision of estimates, but there are many addi¬ 
tional reasons why retaining data streams or creat¬ 
ing new data streams are important. Data availability 
can fluctuate with harvest limits for species for which 
the fishery is the primary data source. Additionally, 
the data collected may be more variable because of 
variations in fishing behavior among fishermen, and 
the data typically will be available only for mature, 
larger animals selected by the fishery. The presence of 
consistent survey data for these stocks could improve 
the ability to produce a more robust estimate of stock 
status. Ideally, survey data would provide comparable 
data across time and space for a large portion of size 
and age classes for a population when it is collected 
by using standardized sampling protocols. Traditional 
trawl survey methods commonly used off the U.S. west 
coast have failed to capture sufficient samples for some 
rockfish species because of gear or area restrictions. 
Creating and maintaining alternative survey sampling 
methods (e.g., hook and line or underwater camera 
sampling) that sample representative portions of a 
stock would be one way to improve the assessment of 
certain rockfish species (e.g., Harms et al, 2008). 
A benefit of continued data collection across mul¬ 
tiple data sources is the potential ability to identify 
misspecification in model assumptions. With the esti¬ 
mation method and operating models applied in this 
study, similar structural assumptions were general¬ 
ly made. However, the true state of nature is never 
known with confidence and continued data collection 
may allow the identification of model misspecification 
in the structural assumptions (e.g., growth, recruit¬ 
ment), allowing models to better approximate reality. 
Specifically, there could be long-term changes in stock 
dynamics that are due to environmental conditions 
(e.g., Hollowed et al., 2011) or biological forces when a 
stock is depleted (e.g., Hixon et al., 2014; Legault and 
Palmer, 2016) that could negatively affect the ability 
of the stock to rebuild. In such a case additional data 
would be required to detect a lack of rebuilding despite 
reduced fishing mortality. Sampling during harvest re¬ 
strictions will provide continued information that can 
identify changes in stock dynamics. Additionally, the 
creation of alternative data streams can buffer against 
the reliance upon a single and potentially variable data 
source and, in turn, could provide valuable insights 
into stock dynamics by the sampling of differing sub¬ 
sections of a population. 
Acknowledgments 
This work has benefitted from comments provided by I. 
Taylor (Northwest Fisheries Science Center), V. Gertse- 
va (Northwest Fisheries Science Center), I. Stewart (In- 
