Austin Frakt posted a roundup of TIE posts on cost sharing, but all his figures either look at out of pocket spending (OOPS) in absolute terms or as a fraction of total health spending, and as I have argued before these figures do not get at the heart of the cost sharing issue. If you want to understand how much people “feel” their health care spending, you should look at data on OOPS as a percent of income. I left a comment over there asking if they knew of any data on this question, but alas I got no response. I’m tired of not having the data so I decided to make my own very rough data series using per capita income data from the census bureau and per capita OOPS data from the OECD. I made graphs using median income and mean income:
As you can see the series are indexed to two different years, which means the actual percentages are going to be wrong, but I don’t think the overall trend is affected by this (Since I don’t know how each series is calculated, there may be other problems unrelated to the index year). Since the OOPS data is indexed to an earlier year than the income data, it means that the OOPS data has been reduced relative to the income data, which means that the true percentages are greater than what is shown – how much greater I don’t know.
Like I said, the actual trend shouldn’t change by changing the index year, so it is possible to say that there has been a definite increase in OOPS as a percent of income except for a short decrease in the 1990’s.
From the graph above it appears that there is both an increase in income and a decrease in OOPS that can explain the dip seen in the 90’s in the first two graphs. So the Clinton boom and the probably the managed care revolution combined to decrease OOPS as a percent of income in the 90’s.
The trend is most likely the result of insurance companies and employers shifting some of the burden of rising health care costs onto the employee. Evidence from consumer directed health care plans has suggested that increased OOPS can moderately slow health spending among the healthy, but we don’t know what the effect of increased OOPS is on system wide health spending. The fact that OOPS as a percent of income has been rising at the same time system wide health spending has been rising is not generally supportive of the cost sharing view, although simply looking at two trends is not hugely informative. It is possible that health spending would have been even greater without the increased OOPS, but like I said looking at two trends is not very informative.
I still wish the US government and all governments would publish official data on OOPS as a percent of income so that we could compare internationally. Official data would be supremely more informative than this makeshift calculation.
New York City is one of the cities where I am really hoping to find a job, and the NY Fed publishes an index of economic activity called the Index of Coincident Economic Indicators for New York City. Recent trends in the Index are very promising:
The index incorporates data on employment, real wages, the unemployment rate, and weekly hours worked in manufacturing. The finance sector of NYC is huge so it shouldn’t be that surprising NYC is recovering given all the government support of the financial sector. Several recent manufacturing surveys have been dismal, and most money market indicators are signaling depressed economic conditions ahead, but hopefully New York can escape that fate.
The other area I’m looking at is Washington DC which is basically recession proof (thank you gubmint).
With a nod to Tyler Cowen, I have signed up at Econ-Jobs at this site: Economist Jobs.
Update: I took down the previous post on high income earners – I want to give it a more thorough treatment.
One of the few “good” things about having been cut off from my University’s research databases is that when I see an interesting gated article in a scholarly journal, I have to set off searching the internet for ungated versions of the paper. When I visit the authors webpage to see if they posted an earlier draft, this sometimes leads me to finding other interesting articles by the same author. Today I was trying to find an ungated version of this article in Health Affairs, and it led me to this paper by Jonathan Kolstad and Amanda Kowalski, “The Impact of Health Care Reform on Hospital and Preventative Care: Evidence from Massachusetts.” Austin Frakt is usually the one to read if you want the latest research on Massachusetts health care reform, and I don’t remember reading about this paper on his site so I thought I would take a look.
The paper sets out to determine what the effect of the Massachusetts health care reform law was on the number of uninsured, access to health care, health care utilization, and hospital costs. It uses data from the CPS and hospital discharge data from the Health Care Cost and Utilization Project National Inpatient Sample, and uses both a difference-in-difference methodology and an Instrumental Variables approach.
Here are some of the main findings:
The number of uninsured among the hospitalized population decreased 2.31 percentage points, and the number of uninsured discharges from hospitals fell 36% from the pre-reform mean. In the hospitalized population there appears to be some crowd-out of private insurance from Medicaid expansions. In the overall population there is little evidence of crowd-out except for a fall in non-group private insurance of .86 percentage points – which is small.
On the extensive margin, the number of hospital discharges (so the amount of people using the hospital) has not changed after the reform compared to other states. On the intensive margin, the length of stay for people in the hospital dropped one percent, and after controlling for the possibility of different patient pools before and after reform the effect is “twice as pronounced.” The authors suggests that the reduction in length of stay could be the result of capacity constraints, but only finds very limited evidence for this.
The authors then study access to preventative care, and examine the admissions through the emergency room; they find that after the reform admissions through the emergency room dropped 5.2%. They find that reductions in ER use is concentrated in lower income people who were less likely to have insurance prior to the reform. This suggests that because people had insurance they were able to seek medical care before their condition got so bad they had to go to the ER.
They study hospitalizations that could have been prevented by appropriate preventative care such as amputations from diabetes or perforations of the appendix due to appendicitis, and find that they were reduced after HCR among people will a less severe condition. Other finds on access to care are:
we see a significant increase of 1.26 percent in individuals reporting they had a personal doctor. The reform also led to a decline in individuals reporting they could not access care due to cost by 3.06 percentage points.
Using 23 different measures of patient safety – which are measures of outcomes that should not occur if appropriate care is given – they find improvements in 13 measures that are statistically and economically significant, no change in 7 indicators and declines in 3.
For hospital costs they provide this chart:
The graphs show that Massachusetts HCR had little effect on the trends in hospital costs, and therefore, on this measure of costs, HCR didn’t bend the cost curve either way. Of course, the Massachusetts reform was not really about bending the cost curve – it was about expanding insurance coverage so this last finding is unsurprising.
Although the Massachusetts HCR and the ACA are not exactly the same, they are similar, and this paper should give pause to anyone who wants to scaremonger about the potential “negative effects” of the ACA by pointing to the Massachusetts experience.
I look forward to future updates of the findings of this paper as more data becomes available since the data necessarily cover a short time period.
Karl Smith looks at the new Industrial Production release today and sees good things. The release says:
Industrial production advanced 0.9 percent in July. Although the index was revised down in April, primarily as a result of a downward revision to the output of utilities, stronger manufacturing output led to upward revisions to production in both May and June. Manufacturing output rose 0.6 percent in July, as the index for motor vehicles and parts jumped 5.2 percent and production elsewhere moved up 0.3 percent. The output of mines advanced 1.1 percent, and the output of utilities increased 2.8 percent, as the extreme heat during the month boosted air conditioning usage. At 94.2 percent of its 2007 average, total industrial production for July was 3.7 percentage points above its year-earlier level.
I decided to take a look at some graphs. First, here is the level of the Industrial Production Index:
The level of production is rising, but it still hasn’t reached it’s pre-recession peak yet which is not surprising given how pathetic the “recovery” has been so far. Here is annualized monthly changes in Industrial Production:
Here is a similar graph showing year-over-year changes in Industrial Production:
Both graphs show a pronounced downward trend in the rate of growth in Industrial Production over the last year. In late 2010 in the Year-Over-Year graph there is a small upward blip probably due to QEII. Karl Smith claimed that Japanese supply disruptions are partly to blame for the downward trend, and in fact, you can see on the Year-Over-Year graph there is a steep(er) fall from March to May 2011 that is probably the result of the Tsunami. At the end of the Year-Over-Year graph over the last few months (including today’s release) the growth rate has blipped up, and hopefully this will mark the end of the downward trend, and the beginning of faster growth in the Industrial Production.
P.S. Wow my graphs are all different sizes.
In a previous post I talked about Karl Smith’s view of automobiles, and decided to take a further look at the data. I’m going to use Real Motor Vehicle Output which contain both cars and truck and is in dollar amounts. You could use unit sales or unit production but it all looks almost exactly alike-except maybe a bit flatter in the mid 2000’s.
What is most striking about the long term data is how closely the rate of increase in motor vehicle (MV) output is to the rate of increase in RGDP. The recent recession was really bad for motor vehicles (obviously, remember “carmageddon”?); the fall in MV output was greater than the overall fall in RGDP. For that reason I think it is unsurprising that the current recovery of MV output has been faster than the recovery in overall output.
Data: BEA
I don’t think this rate of increase can continue; I expect that as the level of MV output approaches its pre-recession level, the rate of increase in MV output will approach the same rate as overall output. With the economy still weak I don’t think MV output will reach its pre-recession trend output. The only way MV output will reach its pre-recession trend level will be for the Fed to act to move the entire economy to its pre-recession trend level. Without Fed action we will be stuck below potential output for a long time.
I was playing around with Google Trends again, and I came across this paper by Hal Varian on “Predicting the Present” with Google trends data. The paper uses a very simple seasonal autoregressive model to predict economic variable with and without Google data. The model they use is:
(1). Log(Variable)= a*Log(Variable T-1) + b*Log(Variable T-1Yr)
(2). Log(Variable)= a*Log(Variable T-1) + b*Log(Variable T-1Yr) + c*Google Trends
Varian argues that using real-time Google Trends data can be useful in predicting official publication of economic data which usually occurs a short while after the time period in question. The paper uses Automotive, Retail, Home Sales and Travel as examples of the usefulness of Google Trends data.
I decided to try this myself with the level of unemployment and Google Insights data on “Unemployment Office.” I ran the regression using data from 2004 through the end of 2010 and then forecasted the 2011 data. Unfortunately, adding the Google Insights data actually increased the Mean Absolute Error of the out of sample forecast from .04 to .05, which is obviously not good. Here is a line graph of the forecasts from Model 1, Model 2, and the actual data:
There are only six out of sample forecasts, which is not a lot, but neither model seems to produce the observed data. So, I have given up on forecasting unemployment, but I got to thinking, “Why would someone google ‘unemployment office’?” and the answer is, “Because they just became unemployed and are looking to get their unemployment insurance.” This would come up in Initial Claims data, so I downloaded it from here. I graphed raw “Unemployment Office” data and Initial Claims data: 
There looks to be a strong correlation between the two so the “Unemployment Office” data will probably be helpful in forecasting initial claims. I ran the two models on data from 2004 through 2010 and forecasted 29 weeks into 2011.
(1a). Log(Initial Claims)= .10+.78*Log(Initial Claims t-1)+.20*log(Initial Claims t-1yr)
(2a). Log(Initial Claims)= 3.05 + .419*Log(Initial Claims t-1) + .318*Log(Initial Claims t-1yr) + .00855*”Unemployment Office)
The Mean Absolute Error without the Google data was .1988, and with the data was .0615, which is a 69% decrease- that is huge. Here is a graph comparing Model 1, Model 2 and the actual data:
The fit using the Google Data (model 2) is significantly better than without the data. I also wanted to see if Google data could help predict the future as well as the present, so I estimated this equation.
(3). Log(Initial Claims)= a*Log(Initial Claims t-1) + b*Log(Initial Claims t-1yr) + c*”Unemployment Office”t-1
I then forecasted the first 29 weeks of 2011 and found a Mean Absolute Error of .1126, which is a 43% decrease in MAE from Model 1. Here is a graph of all three forecasts against actual data from 2011:
It seems that Google data not only improves predictions of the present, but also of the future. This is probably because some people that search for the local unemployment office don’t end up getting there till the following week. It’s kinda cool how search data can be so predictive of how people will actually behave.
A few weeks ago Karl Smith wrote,
However, lets parse out what’s going on here. A major driver is the rise in the price of vehicles. New car price growth continued to be quite strong at .6%, while used car price growth continued to accelerate, all the way to 1.6%.
To put that in perspective 1.6% is an annualized increase of 21%. This is supports a general view that new car sales were operating well below sustainable levels over the course of the past few years. There is a lot of other data to support this, but the United States was simply building too few cars.
This is starting to show up most strongly in a shortage of used cars and hence the strong rise in price. What we expect to see is this push people into the new car market and thus increase industrial production. This is how the connection between inflation and output works in the real world.
At the time he wrote this, I thought there could be an alternate explanation. Since used cars are an inferior good, the demand for used cars will increase when income or expected income fall. Given that consumer sentiment has been falling recently, I felt that that supported my thesis that since people felt poorer they shifted demand from new to used cars and that would explain the price rise in used cars. Our different theses have different predictions-Smith’s implies greater auto production, while mine implies decreased auto production. I didn’t post my thoughts because I didn’t think it explained the 7% annual rise in price of new cars. The new GDP report has findings that can test the prediction of each thesis,
Motor vehicle output subtracted 0.12 percentage point from the second-quarter change in real GDP after adding 1.08 percentage points to the first-quarter change.
This is just one data point, but it does lend some support to my view. It is possible that production could pick up later in the year, but since the Fed has already begun to let its balance sheet start shrinking, and we have reports like this from the WSJ that report people sitting on cash, I expect auto output and overall output will continue its sluggish growth into the second half of the year.
Reports like this from the WSJ about companies and investors taking cash out of markets and sitting it in bank accounts are probably the best example of what David Beckworth calls our excess money demand problem.
Anyone who wants an analysis of News Corp’s handling of the hacking scandal should read Daniel Diermeier because his research interest is in the handling of corporate crises.
First, newspapers and media companies are especially bad at managing reputational crises. This is not only illustrated by the News Corp. crises but other cases such as the Jayson Blair scandal at the New York Times or “Memogate” at CBS’ 60 Minutes. Operating in an intensely politicized atmosphere in their daily lives, news organizations typically interpret any criticism immediately as another politically motivated act and miss the underlying business issue—a crisis that threatens the pillars of news organizations: competence and integrity. The response is defensiveness, which further erodes trust. What was required was a sense of transparency, empathy and commitment to set things right. After almost two weeks, News Corp. finally changed course and took some of the steps that should have been taken much earlier: an apology to the victims of the hacking scandal, the resignation of Rebekah Brooks, and a commitment to reform. But after weeks of fighting back and dismissing the concerns, these steps now look calculated and reactive.
Exogenous Variation 