Analyses of H-1B Salaries in Silicon Valley Firms

Recently Tech Crunch ran an article with the intriguing title “How Google, Facebook And Others Pay Their H-1B Employees,” by Kiran Dhillon.  It was an ambiitious, I might even say daring, attempt, and though its numbers are consistent with my contention that Silicon Valley firms underpay their foreign workers, it was severely flawed.  In this post, I’ll explain the problem with Ms. Dhillon’s analysis, and then show how to fix it, presenting newly updated data from my previous papers. (Note that I’m using the term Silicon Valley as shorthand for famous mainstream high-tech firms, as opposed to the Indian IT services companies.)

I believe you will find these analyses to be valuable, but it does require that you understand how their approach works. Both Dhillon’s analysis and mine center on the fact that H-1B and green card law require that the foreign worker must be paid the prevailing wage, a legal term defined to be the average wage for the given occupation, region and level of experience. Both she and I get our data from the FLC Data Center, which has data both for H-1Bs and green card sponsorees. For reasons that I and others have explained before, the latter data set is more reliable, and that is what I used exclusively in my analyses.

Dhillon reasoned that firms paying much more than prevailing wage are treating their H-1Bs fairly. She identified some that seemed suspect, writing, for instance, “Foreign systems software engineers may want to avoid Microsoft…” In her detailed analysis, she finds that Intel pays “Very Low [Green Card] Salaries: $85,176 Median Salary (18% lower than the median for employers in the same city and industry).”

Given that both Microsoft and Intel have long been in the vanguard of pushing Congress to expand foreign tech worker programs, these are intriguing numbers. However, Dhillon’s numbers cannot be correct. Employers must by law pay at least the prevailing wage (flawed though its legal definition is), so it cannot be true that Intel is paying below that value. The Wage Ratio (WR), i.e. wage earned by the worker divided by the prevailing wage, must be at least 1.00. (Before 2004, a firm could legally pay 5% under prevailing wage, and the data showed many cases in which Intel was doing so.) Judging from Dhillon’s description of her calculation methods, her problem seems to be that she improperly FIRST aggregated wages and prevailing wage figures across occupations, and THEN took the ratio of wages to prevailing wage. In other words, she took a “wage ratio of medians” rather than a “median of wage ratios.”

Moreover, some firms stated their wages in terms of ranges, typically with quite a spread between the two. It’s not clear how Dhillon handled this. My approach has been to limit my analyses to records specifying a single wage. I did this in my Migration Letters and EPI papers, and will present updated data below. But first, it is vital that readers understand the concept, as follows.

The employers claim to hire foreign workers who are “rare” in one of two senses:

  • Skills: They possess “hot” skill sets, e.g. Android programming.
  • Talent: They have high levels of intellectual talent, i.e. are “the best and the brightest.”

Thus the employers should be paying much more than prevailing wage. How much more? Well, I was able to quantify these notions in the above two papers, conservatively at about 20% and 30%, respectively. In other words, if an employer is paying a substantial number of its foreign workers well under, say, 1.20 or 1.30, then either

  • the foreign workers are underpaid or
  • the foreign workers don’t possess “rare” traits after all, so it was not justified to hire them.

Either way, it would suggest that the industry is not being very honest with us. This is what I found in the two papers cited above, based on 2011 data. Let’s see what the situation is today.  Data for 2014 seem to be available only for Quarters 3 and 4; but as the results are similar for all of 2013, let’s look at the 2014 numbers:

Amazon 1.02 0.80 46
Apple 1.00 0.86 42
Cisco 1.00 0.60 5
eBay 1.08 0.38 425
Facebook 1.23 0.00 4
Google 1.19 0.19 1500
Intel 1.76 0.00 1
LinkedIn 1.19 0.08 26
Marvell 1.00 1.00 95
Microsoft 1.03 0.57 14
Oracle 1.11 0.36 11
Twitter 1.27 0.09 149

Again, the small samples for Cisco etc. are due to reporting salary as a range. The 2013 data have the same problem, though with similar results, i.e. one could get a decent sample size for firms like Cisco by aggregating over several years.

To reinforce the premises behind the analysis, look at eBay. Typically it is paying its foreign workers only 8% above prevailing wage, with the word “only” alluding to the fact that this actually represents UNDERpayment of 10-20% below the market worth of the workers.

Google, on the other hand, is typically paying 19% over prevailing wage.  This is much better, but given the firm’s extremely stringent hiring process, it should be getting “the best and the brightest” (and the people I know there, both foreign and domestic, eminently fit that description), Google is UNDERpaying its foreign workers by something like 10%.

The case of Marvell is interesting, in that the founders, a husband and wife team, are from India and China, respectively. It seems to have the worst record among the firms shown above, suggesting that the founders are overly relying on hiring the countrymen. (I’ll give a breakdown by countries in another post.)

In viewing the above data, keep two key points in mind:

  • The really big wage savings accruing from hiring H-1Bs is avoiding hiring the older (35+) Americans, a factor not reflected above.
  • For many of these companies, wage savings is secondary anyway. What they really like about hiring foreign workers is that the latter are immobile; as I’ve explained before, this is viewed as a huge advantage by many tech companies.

One more point: The numbers for the entire F14 Q3/Q4 data are

all firms 1.00 0.70 89499

Remember, this is green card data, and thus basically excludes the Indian IT services firms, who only rarely sponsor their H-1Bs for green cards. In other words, in the mainstream industry as a whole, the foreign workers are typically being paid 20-30% below market. This illustrates the point I’ve made repeatedly: Abuse of the H-1B program pervades the entire industry, not just the Indian bodyshops.

6 thoughts on “Analyses of H-1B Salaries in Silicon Valley Firms

  1. Norm, your point here is strong, that even the “prevailing wage” needs to be categorized because specialized areas pay 20% better. But I remind everyone, that’s what Americans get for the specialty. Any H-1B brought in should be paid *more* than that, for the damage that any import does to the base market, and to help increase (!!!) the rates to solve the underlying shortage, domestically.

    Hey Norm, some column headings on those charts would sure be nice.


  2. > She identified some that seemed suspect, writing, for instance, “Foreign systems software engineers may want to avoid Microsoft…” In her detailed analysis, she finds that Intel pays “Very Low [Green Card] Salaries: $85,176 Median Salary (18% lower than the median for employers in the same city and industry).”

    I played around with replicating her numbers in R and was able to replicate her $85,176 number for Intel and posted the results at . I got that value for all of the applications and for just the certified applications. For green card data, there’s little difference between the two but, for H-1B LCA data, there is much more. For that, it appeared that Dhillon is mistakenly looking at all of the applications though it doesn’t make a big difference. Also, she appears to be ignoring the upper limit of the salary paid (WAGE_OFFERED_TO_9089) though I believe that the lower limit is the more important and is the one that must be higher than the “prevailing wage”. Still, this may explain some of why Intel has a minimal wage paid to prevailing wage ratio of just 1. It appears that they nearly always include a range in their applications.

    As you mentioned, it appears that the green card data is much cleaner than the LCA data. The green card data for the certified versus all applications is pretty close except that Apple did have a maximum wage to prevailing wage ratio of 10 that was not certified. Most importantly, the lowest ratio was 0.899 and it was not certified. However, you can see that in the LCA data, Google submitted an application where the wage was about one tenth of the prevailing wage! Needless to say, that was not certified. In any event, even the LCA data looked somewhat cleaner than past years. I’ve posted examples of some of those errors at .


    • The big problem with the LCA data, as many have pointed out, is that they merely record applications for permission to hire an H-1B. The employer need not have an actual worker at hand. I do find that the LCA data generally track well with the real data, but unfortunately that subtlety is hard to explain to the press and Congress. So, I usually stick with the PERM data.


  3. I did just find that there are still a number of errors in the LCA data, just not so much for the major tech firms. As you can see in Table 6 at , Wipro has a certified application to pay a programmer analyst over $200 million per year. Why can’t I find a job like that! It looks as though the year 2013 got tacked on to the beginning. Items 2 through 6 look as though the wage was listed in cents instead of dollars, giving a wage to prevailing wage ratio of just over 100. Many of the items from 7 through 56 may be from an extra zero being added, giving a ratio just over 10. Some of those just under 10 may be application paying just UNDER the prevailing wage but they got certified because an extra zero was “accidentally” added. Interestingly, there are some big salaries there that may be authentic such as items 54 and 55 (President of Lone Star Global) and item 64 (President and CEO of Estee Lauder).

    Surprisingly, I also some astronomical ratios for green card data. They seem to be caused by Prevailing Wages of 90 and less. Do you know if these are codes and not actual prevailing wages. I did make sure that both the actual and prevailing wages had units of “Year”. In any case, tables 5 and 11 do show that, due to the errors, the median tends to be a safer measurement to use.


  4. One thing, and probably the only thing that LCA data can be used for is to show the premeditated murder, oops, displacement of American Workers in America.

    It is great for that.
    As an example, if a person were to monitor these applications on a city by city basis, you could come pretty close to predicting which city will be our next south california edison or disney.

    That is the main reason I built these maps and provided various options for digging deep into them.

    I would like to find a way to tie these maps to actual contracts by actual government organizations using tax payer revenue to displace Americans in America similar to the california unemployment fiasco that recently made the headlines.

    While we might not be able to stop some things, when you as a government organization use tax payer revenue that was paid in to help the Americans in your state, and you use this money to provide jobs for temporary visa holders rather than the unemployed in your state, well I believe we can shame the hell out of you if we shine a bright enough spotlight on your activities.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s