Full version: Are manufacturing jobs growing or declining? Here’s what the data shows

Background

There have been claims of rapid growth in Indian manufacturing employment in recent years. The growth is said to be both in the factory sector and non-factory sector comprising unregistered manufacturing. On the other hand, while the recent labour force surveys have shown improvement in work participation, especially for women, there have been concerns on the quality of the new employment reported in these surveys. 

The data that forms the basis for these contradictory outlooks come from the well-established and time-tested surveys of the National Statistical Office (NSO). These surveys are the current series of the Periodic Labour Force Surveys (PLFS), the Annual Survey of Industries (ASI) and the Annual Survey of Unincorporated Sector Enterprises (ASUSE).

The PLFS is available from 2017-18, and we have the quinquennial employment–unemployment surveys for the earlier years which are comparable with it. Similarly, the ASUSE is a continuation of the earlier surveys of unorganised sector enterprises that the National Sample Survey Office (NSSO) has been doing regularly, usually at an interval of five years. The ASI, covering registered factories, has been in existence since the 1960s.

The current debates use the datasets mostly from these three sources. Prof B Goldar and Prof Suresh Chand Aggarwal, using these official sources, have concluded that manufacturing employment has grown significantly between 2015-16 and 2023, both in the registered factory segment and the unregistered segment, which is dominated by informal manufacturing. According to them, in the last four years, employment in Indian manufacturing has grown at an average annual rate of about seven percent. Last year, aggregate employment grew by eight percent and manufacturing employment by 10 percent. Rural women accounted for over 50 percent of the jobs created in manufacturing in 2023. Their calculations show the number of workers in manufacturing as 70.5 million in 2022-23. These numbers look quite reassuring in the face of the persistently pessimistic unofficial views of the employment situation in the country. 

What characterises the recent period covered in these surveys is the Covid-19 pandemic’s impact on economic activity during 2020-21 and 2021-22 and its recovery. Another significant event to have impacted this sector is the demonetisation of 86 percent of the currency in circulation in November 2016 which virtually halted economic activities in the country for many months. The third factor is the implementation of the Goods and Services Tax (GST) nationwide from July 2017 that could have impacted the entrepreneurial activity, though its impact on the unorganised sector may not be as significant as demonetisation or the Covid-19 lockdown. 

The PLFS shows that the employment scenario recovered after 2017-18 and clocked 9.3 percent growth in the next year. Growth went down subsequently due to Covid-19 to 0.3 percent in 2021-22. The employment growth during 2022-23 is estimated to be 4.4 percent.

It is in this backdrop that we look at the data from these three sources, including the extent to which these datasets can be integrated. It is also to be noted that the available datasets do not have strict matching reference years, constraining comparisons besides the difference in survey reference periods like for example, the ASI follows the accounting year while the NSSO surveys use the traditional agricultural year July to June.

We start with a brief description of the framework used for measuring employment in these three data sources. In the next section we look at the trends in the growth of number of enterprises and employment in the recent period including changes in employment among industry groups. Subsequently, we briefly look at the nature of employment in manufacturing from the PLFS to understand the type and quality of employment growth.

Conceptual disconnect of employment data in different sources

1. Employment in ASI

The ASI is intended to provide data to understand the changes in the growth, composition and structure of the organised manufacturing sector comprising activities relating to manufacturing, repair services, gas and water supply, and cold storage. The annual survey is conducted using a list of units registered under specified statutes. As is common with all lists, it has a certain element of imperfection. While new units are added as and when reported by the Inspector of Factories or other specified authorities, a closed or non-producing unit does not instantly disappear from the list. 

The annual increase in the number of factories has been somewhat modest, though in 2010-11 there was a huge jump of around 50,000 units. The 2021-22 data shows that there are about 250,000 factories in the list used for the survey. There are indications that starting from 2014-15, the coverage of ASI has been extended beyond the relevant sections of the Factories Act, by adding units with 100 or more employees, not registered under the Factories Act but registered under specified Acts /Boards / Authorities. These are units included in the Business Register of Establishments prepared by state governments and verified by the Field Operations Division (FOD) of the NSSO. It is not clear if this has brought any significant change in the coverage. Although the scope of the ASI covers some units outside manufacturing, we consider only units strictly falling under manufacturing as per the National Industry Classification (NIC 2008) covering industry divisions 10 to 33.

Further, the ASI is not specifically designed to collect data on manufacturing employment by identifying individual workers. It relies on statutory records for the purpose of obtaining data. It records the man-days worked on manufacturing days of the factory. The total number of man-days worked during the accounting year by each category of employees is obtained by adding the number of workers attending in each shift across all shifts worked on all working days during the accounting year. This figure excludes people who are paid but remain on leave/ strike etc. It also provides man-days worked on non-manufacturing days. The man-days worked on repair and maintenance and/or construction activities are also recorded. For factories that have not operated during the year, man-days are reported under non-manufacturing only. The number of people employed in the ASI schedule is arrived at after dividing the total number of man-days worked by the number of days the factory operated.

Thus, the employment data provided in the ASI is tailored to provide an accurate measure of the labour input. Therefore, the number of persons engaged by factories presented in ASI reports is not strictly relatable to any set of specific individuals.

The reference period for the ASI is generally the accounting year of the factory ending on any day during the respective financial year. Therefore, it is not uncommon to have some factory reporting data for more than a year due to extended finalisation of their accounts. Apart from the factories in operation, the ASI frame comprises factories which are categorised as ‘existing with fixed assets and maintaining staff but not having production’ and ‘existing with fixed assets but not maintaining staff and not having production’ etc.

As cautioned in the ASI report, in certain cases, there may be abnormal growth or decline in some characteristics compared to the previous year. The possible reasons for the same could be inclusion of new units in the frame, particularly in the census sector or selection/non-selection of some sample units, having considerable value of multiplier attached to it; high/low performance of the units in the current year, as compared to its performance in the previous year due to various reasons; closure/non-operation of units, which had a significant effect on the economy in the previous year; movement of sample units in the previous year to census units in the current year, depending either on the size of employment or due to the sampling strategy or both, and vice-versa etc. 

2. NSSO surveys of the unorganised sector

The recently released report of the NSSO on the unorganised sector enterprises is part of a series of surveys to understand the contribution of the unorganised sector in the economy. The worker data gathered in these surveys includes both full-time and part-time workers. The categories of workers recorded are working owner, hired worker (both formal and informal), helper/apprentice and other workers. 

For this survey, a worker is a person working within the premises of the enterprise and on its payroll as also the working owners and unpaid family workers. A worker may serve the enterprise in any capacity. A worker need not mean that the same person will be working continuously; it will only refer to a position. i.e., if one person is terminated and another person is appointed in his place, it will be counted as a single worker. A worker engaged for more than half of the normal working hours of the enterprise will be treated as a full-time worker. A part-time worker is a person who works for less than or equal to half of the period of normal working hours of the enterprise on a fairly regular basis.

The estimates of the number of enterprises or employment in these surveys are based on the information collected from selected samples which are areal units. For example, the Sixth Economic Census conducted during 2013-14 enumerated 18.03 million trading enterprises. The unorganised sector survey of 2015-16 estimated the number of trading enterprises as 23.04 million (Table 1). In the manufacturing sector, the Economic Census count of 10.3 million is about half of the estimated number in the unorganised sector (19.7 million), possibly suggesting that the kind of underestimation we observe in the NSSO’s estimate of population is not present in the enterprise surveys.

A strict comparison of the absolute numbers from different types of surveys is not proposed. The number of workers estimated from either the ASI or the ASUSE has a perspective different from a labour force survey covering individuals from households. Foremost among is that there is little scope for counting a person more than once in the household surveys. Any attempt to use the worker data from enterprise surveys and household surveys interchangeably would imply overlooking the basic concepts used in household survey-based employment measures and those from enterprise surveys. The procedure of subtracting ASI-based worker numbers from the PLFS estimates of workers to derive the residual manufacturing workforce should be seen in this context.

3. Data from labour force surveys

Features of employment data from labour force surveys are well known. These surveys have data using different concepts and collect many more details that allow for a much better understanding of the employment situation. When it comes to counting workers, the usual status of persons considering both the principal and subsidiary status is generally used. While the principal status is the status pursued for a major part of the year, the subsidiary status workers would be those who were not working in the principal status but had pursued some work for some time (a minimum of one month is now prescribed). 

Notwithstanding the known limitations of the household surveys, the various rates and ratios estimated from these surveys have been found to be very robust. The fact that the absolute numbers estimated from the surveys are not suitable to be used directly has meant that users apply these ratios on the projected population corresponding to the mid-year of the survey period to obtain national level estimates of workers and unemployed people, etc.

However, the dependence on population projections based on derived/assumed rates of fertility, mortality, migration and rural/urban share etc is also somewhat problematic in the absence of any recent population census. The most authoritative official projections by the National Population Commission of the Ministry of Health and Family Welfare (MoHFW) have also been questioned due to the lower total fertility rate (TFR) reported in the most recent National Family Health Survey (NFHS) compared to what was assumed in these projections. Thus, drawing conclusions of employment numbers in manufacturing that are anchored on population projections are questionable. 

One option would have been to use the estimated workers from two NSSO surveys that have used similar sample designs and sampling frames for sample selection. One may argue that the often-cited usual underestimation of the absolute numbers is expected to be along the same directions in such surveys. Unfortunately, this is not found to be so in the population estimates derived from the PLFS. The usual practice is to look at the rates for components of the population like rural and urban and by gender. One can look at the distribution of workers by industry groups assuming the underestimation/overestimation would be uniform across all categories. Combining the rural and urban estimates using survey population as weights may not be strictly appropriate due to the population underestimation in the urban sector being higher than in the rural population. A better procedure would be to use rural and urban population projections separately on the relevant rates and then aggregate to get the total. We follow this procedure.

Growth in the number of enterprises in manufacturing

1. Estimated number of enterprises in unorganised manufacturing

Our understanding of the size and contribution of activities in the unorganised sector is mainly based on nationwide sample surveys of the NSSO. These surveys are carried out generally once in five years, though the coverage of sectors could be slightly different. After the last two surveys in 2010-11 and 2015-16, the NSSO initiated the ASUSE. Unlike the ASI, which is based on the list of registered units, the NSSO surveys are area-based. All households/structures in the selected domain – either a village or an urban block – are visited to identify entrepreneurial activity. All activities taking place in the fixed locations within the domain and activities done by households without any fixed locations are listed for drawing the samples. However, there are presently no external sources for validating these estimates as these in general are in the informal sector and beyond the pale of government regulations. Nevertheless, there is no reason to doubt the inter-survey comparability of the estimates.

The NSS 67th round (July 2010 – June 2011) covered non-agricultural unincorporated enterprises belonging to three sectors viz., manufacturing, trade and other services excluding construction. Units registered under the Factories Act and those covered under the ASI were excluded. Also excluded were enterprises which are incorporated i.e. registered under Companies Act, 1956, government and public-sector enterprises and cooperatives. The 73rd round (2015-16) also covered all unincorporated non-agricultural enterprises belonging to manufacturing, trade and other services (excluding construction). Again, these included manufacturing enterprises outside the registered factory sector covered under the ASI scheme; non-captive electric power generation, transmission and distribution by units not registered with the Central Electricity Authority like the small units operating generators on diesel, kerosene to produce electricity; trading enterprises and other services sector enterprises excluding construction. Thus, construction is the most significant non-agricultural activity excluded in this survey as was done in 2010-11. The addition to the coverage is the non-captive electric power generation by unregistered units. The additional units covered in the 73^rd round would be small-scale electricity producing units which would be rather small in number. There were minor methodological improvements in the value-added and other economic variables, but these would not impact the estimate of the number of enterprises or the reported workers in them, especially when we consider only the specified industries. 

In the ASUSE of 2021-22 and 2022-23 also, the coverage is similar to the earlier surveys. However, the surveys during 2021-22 had to take into account the field issues due to the Covid-19 restrictions. It would be thus prudent to exclude 2021-22 for comparison purposes.

The overall picture of the unorganised sector, disregarding the coverage changes, presented in the table below, shows that unorganised manufacturing has not performed well in recent years. While unorganised manufacturing reported significant growth both in the number of enterprises and employment from 2010-11 to 2015-16, it recorded a fall in the more recent period. In addition, this data show that the services sector has done well while the number of enterprises in trade has declined. Employment levels in trade are almost at the same level as before.

In the subsequent discussions we concentrate only on the manufacturing part of the survey. From the tables we can see that over 80 percent of the unorganised manufacturing enterprises are in five groups: apparel making (41.1 percent), tobacco products (11.7 percent), textiles (10.5 percent), food products (12.2 percent), and wood and furniture (7.4 percent). 

In Figure 2a, the estimated number of enterprises for major unorganised manufacturing activities for the period 2015-16 and 2022-23 are given. The respective shares of these industries in the total unorganised manufacturing sector are also indicated. Changing shares of the activities indicate the structural shifts in unorganised manufacturing. 

The number of enterprises in the manufacture of wearing apparel rose by 30.7 percent and the share of this activity among the surveyed units went up from 28.5 percent to 41.1 percent. Wearing apparel includes ‘custom tailoring’, which is a common activity done mostly within the precincts of the household and to a lesser extent outside. In fact, the survey shows that 83.5 percent of these enterprises operate from within the household. An analysis of the unit-level data from the PLFS shows that 70 percent of the workers are either own account or unpaid family workers in this industry. Regular salaried workers are only 23 percent and casual labour accounts for 3.8 percent. Only 2.5 percent are in the employer category hiring labour in the enterprise. Such a large share of own account workers is next only to the tobacco products group, where 93 percent are own account workers or in the unpaid category.

The next large activity is the manufacture of tobacco products. Their number has registered a decline of 36 percent. However, the putting out system followed in beedi making could mean that all this activity may not get listed in the enterprise surveys, but would be recorded in the household labour force surveys.

Textile manufacturing also shows a decline. This can possibly be attributed to the decline in home-based manufacturing of textiles like handloom and their shifting to more organised mechanised production units. All other sectors that had a share of over one percent in the total unorganised manufacturing show a decline. 

While the decline in manufacture of tobacco products is in keeping with the regulations and policy, the decline in other main sectors needs further reasoning. The most probable reason in the decline of smaller enterprises in traditional manufacturing could be that these activities have moved to the organised sector due to the overall change in production strategies. For example, manufacturing of food products could have moved to more branded producers from small scale or household enterprises. The small-scale wood-based manufacture is now being done in factories or demand for wood products has come down. The disappearance of the unorganised sector units indicated by the 2022-23 survey could have happened due to their unviability, changing preference of the people for branded or better manufactured items, or failure to stand up to competition from registered units. 

2. Number of manufacturing units in ASI

Unlike unorganised manufacturing, which has a large number of enterprises confined to a few activities, the ASI coverage shows units in most of the activities. In 2015-16, units manufacturing food, non-metallic mineral products, textiles, metal products, rubber and plastic products accounted for over 50 percent of the manufacturing units. One can see that the structure of manufacturing in the organised ASI sector and in the unorganised informal sector is quite different. Therefore, it would not be appropriate to compare the performance in output of employment of these two segments of manufacturing side by side. Certain industries though fewer in number can account for substantial employment. However, the household labour force surveys would cover all types of employment.

In general, we see that the number of factories went up from 2015-16 to 2021-22 for all industries except for tobacco-related manufacturing and for printing, repair and installation industries (not shown in the table). The number of factories manufacturing wearing apparel increased by 19.3 percent. The industry ‘other non-metallic mineral products’ would include mostly brick kilns, which are highly labour-intensive. Tobacco products manufacturing registered a decline in the number of factories by 15.9 percent. Textile manufacturing showed an increase of over two percent. The number of factories involved in wood, furniture and textiles all showed an increase, in contrast to the decline in their unorganised counterparts. However, considering that the period covered is six years, the growth is not very significant.

As noted earlier, it may not be prudent to credit changes in the number of factories straightway to any resurgence in manufacturing. The units in the ASI cannot appear and disappear as easily as the unorganised sector enterprises do. Registered factories require several legal formalities to start manufacturing and their closure is also regulated. But their production capacity can go up, which in most cases would show up in the number of workers and in the value-added figures.

The ASI also reports the number of factories in operation during the year. Looking at the number of factories operating in recent years for the main industry groups, we do find that the number has not significantly increased in recent years, except for the industry group ‘apparel making’. Here, the number of operating factories has gone up from 7,193 in 2015-16 to 9,007 in 2021-22. In some sectors the number has actually decreased in 2021-22. While unorganised manufacturing has indicated a serious decline since 2015-16, the registered sector does not show any significant decline, but the number of operating factories has also not grown in any significant numbers. We do find a significant change in the number of enterprises in a few manufacturing activities, but these activities are mostly those that take place more at the household level.

Employment in manufacturing

In this section, we look at the employment growth reported in manufacturing in these three different sources without making any efforts to combine them. While the number of enterprises is available in ASUSE and ASI, the PLFS provides data on workers engaged in manufacturing from the supply side–the households. Here again we do not have exactly matching periods for making comparisons. But the directions of change in employment in recent years can be analysed.

1. Overall employment growth

The average growth in employment considering all economic activities during 2017-18 and 2022-23 is close to four percent, with rural employment growth being higher. Manufacturing employment has grown at 5.2 percent in rural areas and 2.3 percent in urban areas, giving an overall growth of 3.5 percent. Employment appears to have recovered from the Covid-19 impact.

It may be recalled that the first PLFS in 2017-18 had reported a drastic decline in the female labour force participation rate (FLPR) in rural areas to 17.5 percent from 24.8 percent in 2011-12. For men in rural areas, the rate fell from 54.3 percent to 51.7 percent. The FLPR was 34 percent in 1983 and had been falling since then. This had brought the overall rural worker participation rate (WPR) to a low of 35 percent in 2017-18 from 44 percent in 2004-05. Given this low base the WPR has picked up for most sectors in subsequent years. For major sectors like construction, trade, and hotels and restaurants the growth rate is quite significant. The employment growth in agriculture has been largely from the increased participation of women. During this period, the manufacturing sector has also shown an average growth of over three percent.

2. Growth in manufacturing employment in recent years

Given that the share of workers in manufacturing is much higher in urban areas, we have estimated the manufacturing workers in rural and urban areas separately and aggregated to get the annual figure at the all-India level. Thus, the survey estimates are applied on the projected population as on the mid-point of the survey period. If we consider the six-year period from 2011-12 to 2017-18, we find the employment growth negative for the rural sector and just one percent annual growth for the urban sector. It may be recalled that the first PLFS corresponding to 2017-18 had reported very high unemployment rate and an overall decline in the work participation rates, especially for women. Subsequent PLFS did show an improvement in the overall employment situation and a drastic improvement in the women’s work participation. 

If we further analyse the higher growth rate in manufacturing employment in the last three years, we find that growth has been much higher for women except during the Covid-19 period when manufacturing employment among urban men showed a higher growth rate than in urban women. 

The increase in women’s employment in manufacturing merits an examination of the nature and type of work they are doing. This is important given that women work for less remuneration and most often in a subsidiary capacity. The table below looks at the share of self-employed, unpaid family workers and those in subsidiary status to the total usually employed.

The percentage of women in rural areas reporting self-employment in manufacturing was over three-fourth and is going up. As per the PLFS 2022-23, 83 percent of rural women are in self-employed capacity. Though it is lower in the urban sector, the share is increasing. Clearly, the improvement in employment of women is in the self-employed category. A similar trend is seen in the share of unpaid women workers as well as those working only in a subsidiary capacity.

It would appear from the survey data that growth in employment is significantly in categories that are considered of poor quality.

3. Employment growth in specific industries

Given that employment of women is mostly outside the scope of wage labour, one would expect the industries employing them would be engaged mostly in certain specific type of activities requiring low-end skills and investment. The recent trends will also identify industries either driving the employment growth or dragging it down. The fact that the period covered in the surveys witnessed events like demonetisation, GST implementation and the Covid-19 pandemic would have contributed to the fluctuations. The somewhat inexplicable changes in annual employment in different industries could be partly due to this. For example, employment numbers in the tobacco industry decreased during the Covid-19 period but have recovered subsequently.

Overall, the manufacture of wearing apparel, textiles, tobacco and foods accounts for over 50 percent of the workforce. For women, this is over three-fourth. Clearly, the role of women in manufacturing has a more restricted space. Of the women employed in these four industries, 80 percent are own account-workers or unpaid family workers. 

The manufacture of wearing apparel accounted for 21.5 percent of all the workers in manufacturing in 2022-23. Employment in this sector grew by 17.3 percent during 2021-22 to 2022-23. Textiles, accounting for 10.7 percent of the total employment, showed negative growth in the last year for which we have the data. The employment growth has been rather volatile for this industry in recent years. 

Fabricated metal, furniture making and pharmaceuticals are among the few industries that have shown positive growth all through. But the employment share of these three industries is only around 12 percent of the total manufacturing employment. 

The growth in manufacturing workers was 8.2 percent in 2021-22, possibly accounting for the resurgence after the pandemic. After very low growth during 2018-19 and 2019-20, growth in employment accelerated to 4.7 percent in 2020-21. One cannot discount the effect of Covid-19 as also the field issues in the surveys during this period. The most recent year (2022-23) has shown a decent growth of 4.4 percent, which is close to the employment growth in the economy as a whole.

In Table 10, we look at growth in manufacturing employment from the ASUSE, PLFS and the ASI. The ASUSE covers a seven-year period from 2015-16 to 2022-23. For comparison, we cover only the PLFS 2017-18 to 2021-22 as the corresponding data from the ASI is for this period. These different periods somewhat constrain the discussion of average growth in employment in different industries. As the ASUSE has become an annual feature, it should be possible to make more meaningful comparisons in the future.

Looking at the average growth in employment reported in these three sources, it is clear that ASUSE employment during 2015-16 and 2022-23 is the lowest. Except for apparel making, basic metals, paper products and coke and refining products, all other industries recorded a decline. The most likely reason would be that this period covers the demonetisation and the Covid-19 pandemic, both of which would impact small and informal manufacturing. The 2015-16 survey had shown excellent growth in unorganised manufacturing during the 2011-12 to 2015-16 period, but this growth appears to have been brought down by the subsequent economic disruptions and has not yet reached the 2015-16 level even now. In the case of apparel making, the subsequent period PLFS shows a higher growth and the inclusion of 2022-23 indicates an even greater growth.

Textile manufacturing reported a drop in employment except for the ASI 2015-16 to 2021-22 period, which means it had a much lower base in 2015-16. For manufacture of food products, the ASUSE shows a decline while the PLFS and ASI have positive growth in employment.

Overall, employment in manufacturing in recent years, starting with the post-demonetisation period (except the ASUSE), shows a positive growth of comparable dimensions. The employment growth in the ASUSE has its base in 2015-16 before the demonetisation and hence has a significantly high base to start with. Consequently, the growth rate in the recent survey using the last survey is negative. The employment data of the PLFS and the ASI start after the demonetisation with a lower base. As noted earlier, the ASI employment data will not be much affected due to demonetisation. That the ASI employment growth is lower than the PLFS rate would also indicate that incremental employment has been higher outside the registered manufacturing sectors.

Conclusions

The three independent surveys covering the manufacturing sector we have discussed are part of the regular system of surveys instituted by the government. The three surveys collect data using different conceptual frameworks. These have to be seen and used separately keeping in mind the differences in definitions, reference periods and methodology. Their continued availability will help researchers settle the questions of data comparability we have highlighted in this paper.

The PLFS data show that the employment rates in general went up in 2022-23. This was quite significant for women. Among the major sectors that clocked significantly high growth during the period 2017-18 to 2022-23 included construction, trade, health etc. Growth rates in the primary sectors and manufacturing are similar for this period.

In short, manufacturing employment has grown but not to the extent quoted by some experts. The picture emerging from the three different sources is consistent, but a clear-cut industry pattern is difficult to pinpoint due to the major economic disruptions and its differential effect on industries. The sectors contributing to growth in manufacturing employment are mostly labour-intensive sectors like wearing apparel, food products, etc. The fact that higher growth in women’s employment goes hand in hand with the increasing share of unpaid and subsidiary work and the overwhelming share of self-employment in manufacturing are matters of concern. 

Another observation on the increase in women’s work participation is the decline in those reporting their activity as ‘attended domestic duties only’. One might argue that the special accent on provision of piped water, toilets and cooking gas in rural households has helped them reduce time spent on these activities and take up economic activities. Although these are activities in general don’t require high-end skills, this shift is to be welcomed as a positive outcome of these initiatives.

However, the available data do not show much evidence of any major boost in manufacturing employment either in the organised sector or in the unorganised sector. A clearer picture will emerge when the new ASUSE and PLFS datasets become available for 2023-24. The use of outdated census data for drawing samples for the surveys and the methodology of population projections and their use in estimating worker numbers from survey estimates can also lead to differing conclusions.

Lastly, the sample size in the PLFS or the ASUSE does not permit us to study the incremental changes in the employment situation in a more disaggregated fashion like gender and industry at the state level.