The raw water consumption data set contains 2,599,862 observations and 308,529 households. Often the utility will make billing adjustments by manually changing a household’s water consumption in a particular month to either credit or charge an account. Hence, negative and extremely large consumption values are either errors or billing adjustments and do not reflect actual consumption. Accordingly, we drop annual consumption below the 1st percentile of consumption and above the 99th percentile of consumption, as these are very likely to be outliers that do not reflect actual consumption. For our main specification, we also drop houses that were sold anytime during the current summer months, the current non-summer months, or the lagged summer months (196,925 observations). We also do not have sales price information for all peer groups. For example, if no houses were sold within the peer group for a particular time period. This drop an additional (589,383 observations). Finally, including household-level fixed effects leads to 27,997 singleton observations. The remaining data set, after dropping parcels that had a transaction in the last year, has 1,535,545 observations and 260,307 households. The raw fitted landscaping data set contains 544,882 observations and 74,112 households. Again, we drop observations below the 1st percentile of consumption and above the 99th percentile of consumption for similar reasons as above. The remaining data set contains 540,451 observations and 72,007 households. Finally, including household-level fixed effects leads to 7,313 singleton observations, resulting in a final data set of 531,650 observations and 71,477 households.

中文翻译：

---

居民用水保护中的同伴效应：来自移民的证据

原水消耗数据集包含 2,599,862 个观测值和 308,529 个家庭。通常，公用事业公司会通过手动更改特定月份的家庭用水量来进行计费调整，以贷记或记帐。因此，负的和极大的消耗值要么是错误，要么是帐单调整，并不反映实际消耗。因此，我们将年消费量降至消费的第 1 个百分点以下和消费的第 99 个百分点以上，因为这些很可能是不反映实际消费的异常值。对于我们的主要规范，我们还会删除在当前夏季月份、当前非夏季月份或滞后夏季月份（196,925 个观测值）期间任何时候出售的房屋。我们也没有所有同行群体的销售价格信息。例如，如果在特定时间段内没有房屋在同龄人中出售。这又下降了（589,383 个观察值）。最后，包括家庭级别的固定效应会导致 27,997 个单例观察。剩下的数据集，在丢弃去年有交易的包裹后，有 1,535,545 个观测值和 260,307 个家庭。原始拟合景观数据集包含 544,882 个观测值和 74,112 个家庭。同样，出于与上述类似的原因，我们将观察值降低到消费的第 1 个百分点以下和消费的第 99 个百分点以上。剩余的数据集包含 540,451 个观测值和 72,007 个家庭。最后，包括家庭层面的固定效应会导致 7,313 个单例观察，从而产生包含 531,650 个观察和 71,477 个家庭的最终数据集。