2011年-世界发展银行全球_Reliability_of_Recall_in_Agricultural_Data_29页_1mb

Summary: Reliability of Recall in Agricultural Data

Core Content

This working paper investigates the reliability of recall in agricultural data collected through household surveys in Sub-Saharan Africa. The study focuses on three countries—Kenya, Malawi, and Rwanda—and analyzes the impact of recall periods on the accuracy of data regarding harvest quantities, cash crop sales, and input use (such as fertilizer and hired labor). The main goal is to assess whether recall bias significantly affects the quality of agricultural data collected over extended periods.

Main Points and Key Findings

• Agriculture's Role in Sub-Saharan Africa:
  Farming remains a critical source of income and food for most rural households in Sub-Saharan Africa. In the RIGA study, over 50% of rural households in these countries rely heavily on on-farm income.

• Recall Periods in Surveys:
  Agricultural data are typically collected by asking respondents to recall past events, often over several months. This method introduces the risk of recall bias, where memory fades over time, leading to under or over-reporting of activities.

• Types of Recall Bias:
  The literature identifies three main types of recall error:

  • Telescoping: Misreporting the timing of events.
  • Heaping: Grouping responses into round numbers (e.g., 50 kg bags).
  • Recall decay: Forgetting details of events.

  The paper primarily focuses on recall decay, as it relates to the accuracy of event details over time.

• Impact of Recall Period:
  The study finds little evidence of significant recall bias in agricultural data. Specifically, it shows that:

  • Farmers’ reports of harvest, crop sales, and input use are not significantly different when collected more than 8 months after the event versus immediately after the harvest.
  • More salient events (e.g., cash crops) are less subject to recall decay than less salient ones (e.g., staple crops).
  • Recall decay is more pronounced for labor usage than for fertilizer use, as labor is spread over time while fertilizer is a one-time event.
  • Cash crops are expected to be more reliable in recall due to their economic significance, but the study cautions that this may not always be the case due to multiple sales over time.

• Data and Methodology:
  The study uses three nationally representative household surveys:

  • Malawi Integrated Household Survey (IHS) (2004/2005)
  • Kenya Integrated Household Budget Survey (KIHBS) (2005/2006)
  • Rwanda Enquête Integrale sur les Conditions de Vie des Menages (EICV) (2001)

  These surveys were conducted over a 12-month period, allowing for the analysis of recall periods that vary from 0–3 months to 8–11 months after the harvest. A binary variable was used to classify recall periods, excluding those between 4–7 months from harvest.

  The authors examine the relationship between recall period and reporting accuracy using the following regression model:
  $$
  Y _ {h} = \alpha + \beta T _ {h} + \lambda X _ {h} + \gamma D + \varepsilon_ {h}
  $$
  Where:

  • $Y$ represents outcome variables (harvest, sales, input use).
  • $T$ is a binary variable indicating whether the interview was conducted 8–11 months or 0–3 months after the harvest.
  • $X$ includes household characteristics (landholdings, head gender, age, education, etc.).
  • $D$ includes dummy variables for geographic regions.
  • $\varepsilon$ is the stochastic error term.

• Sample Sizes:
  The final sample sizes vary by crop and country:

  • Maize: 4,435 households in Malawi, 1,018 in Rwanda.
  • Tobacco: 620 households in Malawi.
  • Coffee: 286 households in Kenya, 362 in Rwanda.
  • Sorghum: 633 households in Rwanda.

• Conclusion:
  The findings alleviate concerns about recall bias in agricultural data, particularly for cash crops and fertilizer use. While the study does not claim that all agricultural data are of high quality, it addresses the issue of recall period as a potential source of error, showing that longer recall periods do not significantly affect data accuracy.

Key Information

• Recall Periods:

  • The paper examines recall periods ranging from 0–3 months to 8–11 months.
  • The use of a binary variable for recall periods simplifies analysis but limits the ability to detect nuanced variations in recall accuracy.

• Salience and Recall Accuracy:

  • Salient events (those with economic or social significance) are less likely to be affected by recall decay.
  • Cash crops are considered more salient than staple crops, and thus more reliable in recall.

• Input Use and Recall:

  • Fertilizer use is more accurately recalled than hired labor, as fertilizer is a singular event.
  • Hired labor is more variable and subject to recall errors due to its sporadic nature.

• Methodological Limitations:

  • The use of a binary recall variable limits the ability to assess systematic variations in recall accuracy.
  • The study assumes that harvest months are based on local agronomists' knowledge due to a lack of direct reporting from households or villages.

Implications

The study contributes to the understanding of data quality in agricultural surveys and provides empirical evidence that recall bias is not a major concern for certain types of agricultural data, particularly those related to cash crops and fertilizer use. It highlights the importance of salience in mitigating recall decay and suggests that multi-topic surveys can still provide reliable data on agricultural activities, even when conducted over extended periods.