Page 73 - IJPS-1-1

Raphael J. Nawrotzki, Fernando Riosmena, Lori M. Hunter, and Daniel M. Runfola

ity-level predictors (indicated by subscript ik), and it has been shown that a two-level
model structure is appropriate for such variables (Barber, Murphy, Axinn et al., 2000). All
models control for the effect (β n) of various sociodemographic factors (x n) on the probabil-
ity to migrate. These controls can operate both at the household and municipality levels,
indicated by the generic subscript z.
Although tests have shown that recall bias is of little concern for the MMP data (Massey,
Alarcon, Durand et al., 1987), we included a measure for the survey year to account for
residual recall error. Finally, the parameter u k constitutes the municipality’s random effects
term that accounts for the nesting of households within municipalities. The multi-level
event history models were estimated using the package lme4 (Bates, 2010; Bates,
Maechler, Bolker et al., 2014) within the R statistical environment (RCoreTeam, 2015).
During the 1986–1999 study period, n = 819 households reported undocumented moves
while only n = 95 households reported documented moves. Although a documented move
constituted a rare event, discrete-time event history models are specifically designed for
small numbers. Simulation exercises have demonstrated that at least five events per pre-
dictor are necessary to produce unbiased and reliable estimates (Vittinghoff and
McCulloch, 2007). The fitted models (Table 2) contained 19 substantive predictors, yield-
ing an average of five events per predictor for the total of 95 documented migration events,
which constituted a sufficiently large number to produce valid and stable results.

3. Results

In line with prior work, results from the multi-level event-history models (Table 2) re-
vealed that undocumented migrations most likely occurred from male headed households
without young children in which the household head has little education and work experi-
ence, is employed in a blue collar occupation and does not own a business or property
(Fussell, 2004; Massey, Alarcon, Durand et al., 1987; Massey and Parrado, 1998;
Nawrotzki, Riosmena and Hunter, 2013; Woodruff and Zenteno, 2007). The presence of
migrant networks strongly facilitates both documented and undocumented migrations
(Fussell and Massey, 2004; Massey and Espinosa, 1997). In contrast, documented mi-
grants are usually better educated and come from areas less dependent on agricultural
production (Fussell, 2004). As the primary analytical focus, the models also included the
two climate change indices.
The results show that climate change significantly influenced international migration
from rural Mexico to the U.S. but that this relationship exclusively emerged for undocu-
mented moves. The significant temperature effect suggested that an increase in warm spell
duration by one standard deviation unit increased undocumented international
out-migrations by 19% (Odd Ratio [OR] = 1.19). In contrast, an increase in precipitation
during extremely wet days by one standard deviation reduced the odds of an undocu-
mented international move to the U.S. by 18% (OR = 0.82).

4. Discussion and Conclusions

Combining detailed migration histories with two climate change indices based on daily
temperature and precipitation information, this study provides evidence that rural Mexican
households employed migration as an adaptation strategy in the face of adverse climate
variability and change. However, while the results demonstrate that climate change sig-
nificantly influenced undocumented migrations, it had no impact on documented moves.
As it is often difficult to obtain a valid work visa given the quotas, backlogs and application
costs (Papademetrious and Terrazas, 2009), households may resort to undocumented border

International Journal of Population Studies | 2015, Volume 1, Issue 1 67