May 7

Summary

• For the Ghana SES survey, I couldn’t find a way to find geographic coordinates more precise than the district level. Temperature at the district is fairly imprecise so not sure if it is a meaningful measurement
• In Bangalore, we wanted to confirm that the null result is well measured: the result is consistent for people who experience really hot weather and across those who experience a large range of temperature. However, overall experience of temperature ranges is less than in Ghana.
• I used DHS data in 3 countries where we have PHQ, GAD: Bangladesh, Mozambique and Nepal. We see almost 0 impact of temperature.
  • Note: data is not a panel so we are comparing people within a DHS cluster

From Last Time

Temperature seems to affect mental health in Ghana but not in Bangalore. Existing papers show some impact on mental health so it is strange that we found a precise null in Bangalore. Additionally, the results conflicts with what we see in Ghana.

To explore these further, we are looking to:

• Understand channel for how temperature affects mental health
• Rule out measurement issues in Bangalore data

Ghana

	All Measures - Anderson	PHQ-2 & GAD-2 - Anderson	Perceived Stress	Worry
	(1)	(2)	(3)	(4)
coef
Daily Mean Temperature	0.0165*** (0.0042)	0.0179*** (0.0044)	0.0082* (0.0049)	0.0087* (0.0048)
fe
period	x	x	x	x
pid	x	x	x	x
stats
Observations	16,938	16,938	16,938	16,938
R-squared	0.386	0.312	0.261	0.294
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

Bangalore

	All Measures - Anderson	PHQ + GAD - Anderson	Loneliness - Z-score	Depression - Z-score
	(1)	(2)	(3)	(4)
coef
Daily Mean Temperature	0.0024 (0.0048)	0.0037 (0.0047)	-0.0009 (0.0019)	0.0013 (0.0022)
fe
month	x	x	x	x
pid	x	x	x	x
stats
Observations	6,038	6,043	6,047	6,047
R-squared	0.651	0.648	0.589	0.636
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

Quick

Ghana SES data

We considered using Ghana SES survey to dig deeper into agricultural channel and explore some of the seasonality in mental health. This helps us nail farming as a channel and address correlation between temperature and lean season given that the mental health shows empirical patterns tied to agricultural seasonality.

• We do not have coordinates for the enumeration areas. Finest level of data I could find is district level but it seems districts are quite hetregenous in temperatures

Using district level temperature data seems noisy given the large variation in temperatures within districts. Below are maps showing the temperature variation within a single district and single hour based on satellite data. In some districts, the variation is pretty large, close to 10 C, which makes me think that using temperature data at the district level is too noisy.

Bangalore - Measurement

In Bangalore, we wanted to make sure the null result is not driven by measurement error. As a first check, below show results for a subsample of people who experience hot temperatures (>84F or 29C) and experience a large variance (>10F or >15F).

The first two subsample show no heterogeneity from the main results. The Range >15 subsample seems to have notably different coefficients for PHQ-GAD anderson, but the N is fairly small at 486.

Ever Above 84

	All Measures - Anderson	PHQ + GAD - Anderson	Loneliness - Z-score	Depression - Z-score
	(1)	(2)	(3)	(4)
coef
Daily Mean Temperature	-0.0094 (0.0083)	0.0001 (0.0082)	-0.0062* (0.0034)	-0.0005 (0.0038)
fe
month	x	x	x	x
pid	x	x	x	x
stats
Observations	1,154	1,156	1,156	1,156
R-squared	0.668	0.637	0.623	0.615
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

Range Above 10

	All Measures - Anderson	PHQ + GAD - Anderson	Loneliness - Z-score	Depression - Z-score
	(1)	(2)	(3)	(4)
coef
Daily Mean Temperature	-0.0016 (0.0062)	0.0015 (0.0062)	-0.0028 (0.0026)	0.0019 (0.0028)
fe
month	x	x	x	x
pid	x	x	x	x
stats
Observations	2,403	2,406	2,407	2,407
R-squared	0.651	0.65	0.575	0.633
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

Range Above 15

	All Measures - Anderson	PHQ + GAD - Anderson	Loneliness - Z-score	Depression - Z-score
	(1)	(2)	(3)	(4)
coef
Daily Mean Temperature	-0.0152 (0.0136)	-0.0128 (0.0138)	-0.0031 (0.0052)	-0.0028 (0.0064)
fe
month	x	x	x	x
pid	x	x	x	x
stats
Observations	486	487	487	487
R-squared	0.642	0.601	0.666	0.59
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

However, the people in Ghana do experience more temperature variation in the data; a much higher proportion of people experience more than 10F compared to our sample in Bangalore.

Sample	Number individuals who experience above 84F	Average range experience by individuals	% who experience above 5F range	% who experience above 10F range	% who experience above 15F range
Bangalore	199	8.2F	74.7%	34.8%	6.6%
Ghana	2160	11.2F	99.8%	71.4%	4.9%

Raw mental health score distributions

The Bangalore null may be partly about the longer frame between measurements. Below, the Bangalore raw PHQ-2 and GAD-2 score distributions are shown next to the Ghana short-form scores.

Distribution of raw scores

PHQ-2

GAD-2

DHS

The DHS surveys give us a broader cross-country comparison using Bangladesh, Mozambique, and Nepal. Outcomes are standardized within country; the main specification uses country-cluster fixed effects and daily mean temperature.

Temperature Exposure

Sample	Individuals above 84F	Average range experienced	% above 5F range	% above 10F range	% above 15F range
All Countries	27961	17.4F	99.1%	90.5%	62.2%
Bangladesh	13367	14.9F	98.2%	83.5%	43.1%
Mozambique	11283	18.5F	99.6%	92.9%	70.2%
Nepal	3311	19.7F	100.0%	98.2%	81.1%

Sample	Average observations per cluster	Mean date range in cluster	Max date range in cluster
All Countries	28.8	4.4 days	175.0 days
Bangladesh	29.7	2.0 days	41.0 days
Mozambique	30.1	7.5 days	175.0 days
Nepal	26.0	3.6 days	23.0 days

Regressions

Full Sample

	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson
	(1)	(2)	(3)
coef
Daily Mean Temperature	0.0000 (0.0031)	-0.0007 (0.0032)	-0.0004 (0.0031)
fe
Country-cluster FE	x	x	x
stats
Observations	50,905	50,905	50,905
R-squared	0.143	0.14	0.154
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

Full Sample, Country FE

	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson
	(1)	(2)	(3)
coef
Daily Mean Temperature	-0.0010* (0.0006)	-0.0028*** (0.0006)	-0.0020*** (0.0006)
fe
Country FE	x	x	x
stats
Observations	50,905	50,905	50,905
R-squared	0.000055	0.000464	0.000242
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

By Country

Bangladesh

	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson
	(1)	(2)	(3)
coef
Daily Mean Temperature	-0.0013 (0.0081)	0.0008 (0.0082)	-0.0003 (0.0080)
fe
Cluster FE	x	x	x
stats
Observations	19,987	19,987	19,987
R-squared	0.148	0.137	0.156
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

Mozambique

	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson
	(1)	(2)	(3)
coef
Daily Mean Temperature	0.0040 (0.0040)	0.0001 (0.0040)	0.0022 (0.0040)
fe
Cluster FE	x	x	x
stats
Observations	18,563	18,563	18,563
R-squared	0.159	0.165	0.176
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

Nepal

	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson
	(1)	(2)	(3)
coef
Daily Mean Temperature	-0.0089 (0.0064)	-0.0038 (0.0064)	-0.0068 (0.0065)
fe
Cluster FE	x	x	x
stats
Observations	12,355	12,355	12,355
R-squared	0.111	0.108	0.117
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

Non-linearity

	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson
	(1)	(2)	(3)
coef
Daily Mean Temperature	-0.0494** (0.0219)	-0.0175 (0.0219)	-0.0359* (0.0217)
Daily Mean Temperature Squared	0.0003** (0.0002)	0.0001 (0.0002)	0.0002* (0.0002)
fe
Country-cluster FE	x	x	x
stats
Observations	50,905	50,905	50,905
R-squared	0.143	0.14	0.154
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

Threshold Temperature Measures

	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
coef
Minimum Sleep Temperature Above 70F	-0.0032 (0.0061)	-0.0063 (0.0061)	-0.0051 (0.0060)
Maximum Sleep Temperature Above 70F				0.0011 (0.0027)	0.0003 (0.0027)	0.0007 (0.0027)
Daily Mean Temperature Above 84F							-0.0025 (0.0180)	-0.0136 (0.0180)	-0.0087 (0.0179)
Daily Maximum Temperature Above 84F										0.0036 (0.0030)	0.0003 (0.0030)	0.0021 (0.0029)
fe
Country-cluster FE	x	x	x	x	x	x	x	x	x	x	x	x
stats
Observations	50,905	50,905	50,905	50,905	50,905	50,905	50,905	50,905	50,905	50,905	50,905	50,905
R-squared	0.143	0.14	0.154	0.143	0.14	0.154	0.143	0.14	0.154	0.143	0.14	0.154
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

Alternative Temperature Measures

	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
coef
Daily Maximum Temperature	0.0003 (0.0019)	0.0002 (0.0019)	0.0003 (0.0019)
Minimum Sleep Temperature				-0.0008 (0.0025)	-0.0017 (0.0025)	-0.0013 (0.0025)
Past 2 Weeks Mean Temperature							-0.0186* (0.0110)	-0.0116 (0.0110)	-0.0162 (0.0109)
Past Month Mean Temperature										-0.0171 (0.0122)	0.0017 (0.0122)	-0.0083 (0.0121)
fe
Country-cluster FE	x	x	x	x	x	x	x	x	x	x	x	x
stats
Observations	50,905	50,905	50,905	50,905	50,905	50,905	50,905	50,905	50,905	50,905	50,905	50,905
R-squared	0.143	0.14	0.154	0.143	0.14	0.154	0.143	0.14	0.154	0.143	0.14	0.154
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

Agricultural

	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson
	(1)	(2)	(3)
coef
Daily Mean Temperature	0.0037 (0.0066)	-0.0024 (0.0067)	0.0007 (0.0066)
fe
Country-cluster FE	x	x	x
stats
Observations	13,161	13,161	13,161
R-squared	0.208	0.208	0.217
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)

Non-agricultural

	PHQ - Z-score	GAD - Z-score	PHQ - GAD Anderson
	(1)	(2)	(3)
coef
Daily Mean Temperature	0.0002 (0.0036)	0.0017 (0.0036)	0.0010 (0.0036)
fe
Country-cluster FE	x	x	x
stats
Observations	37,744	37,744	37,744
R-squared	0.172	0.167	0.184
Significance levels: * p < 0.1, ** p < 0.05, *** p < 0.01. Format of coefficient cell: Coefficient (Std. Error)