Does the Future Affect the Present? The Effects of Future Weather on the Current Collection of Planted Crops and Wildlife in a Native Amazonian Society of Bolivia

Abstract

Unlike neighboring disciplines, anthropology rarely studies how actual future events affect current behavior. Such studies could lay the groundwork for studies of ethno-forecasting. Psychologists argue that people forecast poorly, but some empirical work in cultural anthropology suggests that at least with weather, rural people might make reasonably accurate forecasts. Using data from a small-scale, pre-industrial rural society in the Bolivian Amazon, this study estimates the effects of future weather on the current collection of planted crops and wildlife. If actual future events affect current behavior, then this would suggest that people must forecast accurately. Longitudinal data covering 11 consecutive months (10/2002–8/2003, inclusive) from 311 women and 326 men ≥age 14 in 13 villages of a contemporary society of forager-farmers in Bolivia’s Amazon (Tsimane’) are used. Individual fixed-effect panel linear regressions are used to estimate the effect of future weather (mean hourly temperature and total daily rain) over the next 1–7 days from today on the probability of collecting wildlife (game, fish, and feral plants excluding firewood) and planted farm crops (annuals and perennials) today. Daily weather records come from a town next to the Tsimane’ territory and data on foraging and farming come from scans (behavioral spot observations) and surveys of study participants done during scans. Short-term future weather (≤3 days) affected the probability of collecting planted crops and wildlife today, although the effect was greater on the amount of planted crops harvested today than on the amount of wildlife collected today. Future weather beyond 3 days bore no significant association with the amount of planted crops harvested today nor with the amount of wildlife collected today. After controlling for future and past weather, today’s weather (mean hourly temperature, but not rain) affected the probability of collecting wildlife today, but today’s weather (temperature or rain) did not affect the probability of collecting planted crops today. The study supports prior work by anthropologists suggesting that rural people forecast accurately. If future weather affects the probability of harvesting planted crops and collecting wildlife today, then this suggests that Tsimane’ must forecast accurately. We discuss possible reasons for the finding. The study also supports growing evidence from rural areas of low-income nations that rural people tend to protect their food production and food consumption well against small idiosyncratic shocks or, in our case, against ordinary daily weather that is not extreme. However, the greater responsiveness of daily foraging output compared with daily farming output to today’s weather suggests that foraging might not protect food consumption as well as farming against adverse climate perturbations.

Appendices

Appendix 1: Sources of Weather Data and Construction of Weather Variables

A. Sources for Weather Data

• 1) Source: National Oceanic & Atmospheric Administration (NOAA); US department of Commerce

• a) Home page. http://www.noaa.gov/

• b) Data downloaded from this link. http://www.ncdc.noaa.gov/oa/ncdc.html (Data link)

• c) For free data. http://www.ncdc.noaa.gov/oa/mpp/freedata.html

• d) Scroll down to free data J- Surface data- Global summary of the day.

  • Select the country

  • Choose the station of interest (in this case San-Borja, Rurrenabaque, and Trinidad)

• e) Data from Jan 2002 to Dec 2003 was downloaded.

• 2) Address:

  National Climatic Data Center

  Federal Building

  151 Patton Avenue

  Asheville NC 28801-5001

  1-828-271-4800

  FAX: 1-828-271-4876

  Email: ncdc.info@noaa.gov

  All contact information for various departments can be access through this link

  http://www.ncdc.noaa.gov/oa/about/ncdccontacts.html

• 3) Coding

  First record—header record.

  All ensuing records—data records as described below.

  All 9’s in a field (e.g., 99.99 for PRCP) indicates no report or insufficient data.

  FIELD	POSITION	TYPE	DESCRIPTION
  STN	1–6	Int	Station number (WMO/DATSAV3 number) for the location.
  WBAN	8–12	Int	WBAN number where applicable—this is the historical “Weather Bureau Air Force Navy” number—with WBAN being the acronym.
  YEAR	15–18	Int	The year
  MODA	19–22	Int	The month and day
  TEMP	25–30	Real	Mean temperature for the day in degrees Fahrenheit to tenths. Missing = 9999.9 (Celsius to tenths for metric version.)
  MAX	103–108	Real	Maximum temperature reported during the day in Fahrenheit to tenths—time of max temp report varies by country and region, so this will sometimes not be the max for the calendar day.
  			Missing = 9999.9 (Celsius to tenths for metric version.)
  Flag	109–109	Char	Blank indicates max temp was taken from the explicit max temp report and not from the ‘hourly’ data. * indicates max temp was derived from the hourly data (i.e., highest hourly or synoptic-reported temperature)
  MIN	111–116	Real	Minimum temperature reported during the ay in Fahrenheit to tenths—time of min temp report varies by country and region, so this will sometimes not be the min for the calendar day.
  			Missing = 9999.9(Celsius to tenths for metric version.)
  Flag	117–117	Char	Blank indicates min temp was taken from the explicit min temp report and not from the hourly’ data. * indicates min temp was derived from the hourly data (i.e., lowest hourly or synoptic-reported temperature)
  PRCP	119–123	Real	Total precipitation (rain and/or melted snow) reported during the day in inches and hundredths; will usually not end with the midnight observation—i.e., may include latter part of previous day.
  			.00 indicates no measurable precipitation (includes a trace).
  			Missing = 99.99 (For metric version, units = millimeters to tenths and missing = 999.9.
  			Note: Many stations do not report ‘0’ on days with no precipitation—therefore, ‘99.99’ will often appear on these days. Also, for example, a station may only report a 6-h amount for the period during which rain fell. See Flag field for source of data.
  Flag	124–124	Char	A = 1 report of 6-h precipitation amount.
  			B = Summation of 2 reports of 6-h precipitation amount.
  			C = Summation of 3 reports of 6-h precipitation amount.
  			D = Summation of 4 reports of 6-h precipitation amount.
  			E = 1 report of 12-h precipitation amount.
  			F = Summation of 2 reports of 12-h precipitation amount.
  			G = 1 report of 24-h precipitation amount.
  			H = Station reported ‘0’ as the amount for the day (e.g., from 6-h reports), but also reported at least one occurrence of precipitation in hourly observations—this could indicate a trace occurred, but should be considered as incomplete data for the day.
  			I = Station did not report any rain data for the day and did not report any occurrences of precipitation in its hourly observations—it’s still possible that rain occurred but was not reported.

  The NCDC Climate Services Branch (CSB) is responsible for distribution of NCDC products to users. NCDC’s CSB can be contacted via the following phone number, internet address, or fax number:

  Telephone number: 1-828-2714800

  Fax number: 1-828-2714876

  Internet address: ncdc.orders@noaa.gov

• 4) Methods used to impute the missing values

  • Daily temperature comes from hourly records of temperature. Daily temperature was converted to centigrade using the following formula,T in C = [(T in F-32)/9]*5

  • Rain data refers to the total for a given day and was given in inches; multiplied by 2.54 to convert into centimeters

  • If San Borja had a missing value, we imputed the mean value from Trinidad and Rurrenabaque, two nearby towns. Recent publications contain discussion of imputation methods for missing weather data (Godoy et al. 2008a, b).

B. Construction of Weather Variables

• Log transformation. We took the natural logarithm (hereafter log) of daily temperature. We added +1 to daily total rain before taking the log of daily rain to avoid producing missing values for days without rain.

• First step: measures of future weather day by day. We took the log of the mean daily hourly temperature and the log of daily total rain for 1, 2, 3, 4, 5, 6, and 7 days after today. Day 1 after today refers to tomorrow, day 2 after today refers to the day after tomorrow, etc. The first step produced a total of 14 variables for future weather, seven variables capturing daily total rain for each of the next 7 days after today and another seven variables capturing the mean of hourly temperature for each of the next 7 days after today.

• Second step: future weather—mean values. Drawing on the values from the first step we took the mean of the log of daily rain and the mean of the log of daily temperature for seven future periods. For example, we estimated (1) the mean amount of total daily rain (in logs) for tomorrow (day 1), (2) the mean amount of total daily rain (in logs) for tomorrow and the day after tomorrow (mean of day 1 and day 2 after today), or (3) the mean amount of total daily rain (in logs) for the next 7 days from today (mean of days 1 + 2 + 3 + 4 + 5 + 6 + 7 after today). This step produced 14 additional variables for the mean of weather variables for different periods of time in the future (e.g., mean temperature of tomorrow and the day after tomorrow; mean temperature of the next 7 days).

• Third step: measures of past weather day by day. We took the log of the mean daily hourly temperature and the log of daily total rain for 1, 2, 3, 4, 5, 6, and 7 days before today. In the previous sentence, day 1 before today refers to yesterday, day 2 before today refers to the day before yesterday, etc. The third step produced a total of 14 variables for past weather, seven variables capturing daily total rain for each of the previous 7 days before today and another seven variables capturing the mean of hourly temperature for each of the previous 7 days before today.

• Fourth step: past weather—average of last 7 days before today. Drawing on the values from the third step, we took the mean of the log of rain and the mean of the log of temperature for the 7 days before today. The fourth step produced two variables for past weather: the mean daily total rain and the mean daily temperature for the 7 days before today.

• Fifth step: today’s weather. We took the log of today’s total rain and the log of today’s mean hourly temperature.

In the regressions we only use the variables from the second, fourth, and fifth steps; the variables from the first and third step were used as inputs to construct variables in the other steps. All the variables from the fourth and the fifth step appear in all regressions; that is, in all regressions we control for the weather during the previous 7 days before today (fourth step) and for today’s weather (fifth step). Among the variables of the second step (future weather), only some variables are entered in each regression, as shown in the first columns of Tables 2, 3 and 4.

Appendix 2:

A Note on the Sample Size of the Regressions

In Table 2, the sample size of the regressions with tomorrow’s weather as an explanatory variable (row 1) is smaller (n = 7099) than the sample size (n = 7235) of the other regressions (rows 2–7). The difference arises from the way STATA computes the mean of variables across rows. The STATA command “egen x = rowmean (×1 ×2 ×3)” produces the mean of x1, x2, and x3; if the value of one variable is missing, STATA estimates a mean for the remaining two variables. Nine days had missing data for tomorrow’s rain or for tomorrow’s temperature, and for these days the mean rain and the mean temperature for tomorrow were set to missing values. There were no days that had two or more consecutive future days of missing values for weather variables. Therefore, the mean of the weather variables for days 1 + 2, days 1 + 2 + 3, days 1 + 2 + 3 + 4, days 1 + 2 + 3 + 4 + 5, days 1 + 2 + 3 + 4 + 5 + 6, and days 1 + 2 + 3 + 4 + 5 + 6 + 7 into the future had more observations than the variable for only tomorrow’s weather (mean of day 1) since the mean of a weather variable for two or more days into the future always produced a non-missing value, even if one of the days in the future had a missing value. In the article we show that the main results do not change if we exclude the observations with missing data for tomorrow’s weather and run all the regressions with the same number of observations.